CFU vs. MPN: Which Method Should You Use to Count Bacteria?

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-11-10

CFU counting measures bacteria by tallying the colonies that grow on an agar surface, while MPN estimates the concentration statistically from the pattern of positive and negative growth in a dilution series.

The rule of thumb is short. Plate when the expected concentration is high enough to land inside the countable range and the matrix lets a colony be seen and counted. Choose MPN when the level is too low to plate reliably, when the cells are stressed or injured, or when the sample is too turbid or particulate for membrane filtration.

That decision drives real consequences. The two methods do not produce interchangeable numbers, they carry different error structures, and they fail in different ways. A laboratory that plates a sample containing three cells per gram will report zero. A laboratory that runs an MPN on a sample containing ten million cells per gram will report a censored "greater than" value. Neither result is wrong. Both are the correct answer to a badly matched question.

This guide walks through how each method actually works at the bench, where the numbers come from, what the countable range and the MPN table really mean, and how to choose between them for water, food, environmental, and diagnostic samples.

The Core Difference in One Paragraph

A CFU is a colony you can see. An MPN is a probability.

Plate counting converts a physical event into a number by arithmetic. One cell, or one clump of cells that stayed together, divides until a colony becomes visible. You count the colonies, multiply by the dilution factor, and divide by the volume plated. The result is discrete. A count of 37 colonies means 37 colony-forming units were recovered from that plate.

MPN converts a set of binary outcomes into a number by statistical inference. You inoculate replicate tubes at several dilution levels and score each tube as positive (growth, gas, turbidity, or a color change) or negative. The pattern of positives across the dilution series is then matched against a table derived from the Poisson distribution. The result is a maximum-likelihood estimate with an attached confidence interval. No single tube contains exactly 43 organisms. The number 43 is the concentration most consistent with the pattern you observed.

Because of that difference, CFU data are counts and MPN data are estimates. A CFU value carries Poisson counting error. An MPN value carries a wider, asymmetric confidence interval that grows rapidly at the low and high ends of the tabulated range.

Side-by-Side Comparison: CFU vs. MPN

The table is the map. The rest of this article explains the terrain.

Colony-Forming Units: A Direct Count With Hidden Assumptions

How a plate count becomes a number

Three plating formats dominate practice. In a spread plate, a small volume (commonly 0.1 mL) is pipetted onto a dried agar surface and spread evenly with a sterile bent rod. In a pour plate, the sample is mixed with molten agar held near 45 °C and poured into a dish, so colonies develop both on the surface and inside the medium. In membrane filtration, a larger volume (commonly 100 mL of water) is pulled through a 0.45 µm pore filter, and the filter is placed on an agar pad or plate so that colonies grow in a tight, countable plane.

The arithmetic is the same for all three. Colonies counted, divided by the volume plated, multiplied by the reciprocal of the dilution. A spread plate showing 42 colonies from 0.1 mL of a 10⁻³ dilution represents 42 ÷ 0.1 × 10³, or 4.2 × 10⁵ CFU/mL.

That clean multiplication is why CFU counting remains the reference method across food, water, and pharmaceutical microbiology. It is transparent, it is cheap per plate, and any auditor can recompute it from the bench sheet.

The countable range and why it exists

Plates are only quantitative within a defined window. Below roughly 25 to 30 colonies, a single colony changes the answer by several percent, and the Poisson error becomes large relative to the count. Above roughly 250 to 300 colonies, colonies merge, compete for nutrients, and produce a crowded lawn that undercounts the true population.

Standard methods therefore specify a countable range and require that at least one dilution fall inside it. Rules of thumb vary by method and matrix, but the principle does not. If every plate is too crowded, the answer is a lower bound. If every plate is empty, the answer is "less than" the plating limit. A result reported from outside the countable range is an extrapolation, not a measurement.

Why one colony is not one cell

The term colony-forming unit exists precisely because the colony count undercounts cells. Staphylococci grow in grape-like clusters. Streptococci form chains. Mycobacterium cells clump. A single colony may descend from one cell or from dozens that were stuck together when the sample was plated.

This is not a rounding error. It is a systematic bias in one direction, and it means CFU values should always be read as "recoverable colony-forming units" rather than "cells." The same logic explains why CFU counts from a homogenized tissue and from a surface rinse of the same item rarely match exactly.

Recovery also depends on the medium. A selective agar with an antibiotic supplement suppresses competitors but also suppresses sublethally injured target cells. A study of Escherichia albertii on raw ground beef used lactose-free MacConkey agar containing 50 mg/L nalidixic acid and 0.5% L-rhamnose to recover a nalidixic-acid-resistant strain inoculated at a target of 4.0 log CFU/g, a design that depends entirely on the strain's resistance marker rather than on the medium's selectivity alone [1]. Change the strain and the recovery changes.

Precision: what the Poisson distribution tells you

A colony count is a sample from a Poisson process. If the expected number of colonies on a plate is N, the standard deviation of repeated counts is √N and the coefficient of variation is 1/√N.

Three worked values make the point:

• 100 colonies: CV = 10%
• 25 colonies: CV = 20%
• 4 colonies: CV = 50%

This is why the countable range exists. It is not arbitrary. It is the zone where counting error is small enough that the dilution series, not the counting statistics, dominates the uncertainty.

Well-controlled quantitative methods can do much better than raw plating when they are calibrated. A terminal-restriction fragment length polymorphism assay evaluated for absolute and species-specific enumeration of a three-species culture reported a maximum coefficient of variation of 4.39% across defined mixed replicates and 1.76% in a three-species chemostat, with linear quantification from 10⁴ to 10⁶ CFU/mL at the lower working range and up to 10¹⁰ CFU/mL at the upper end [2]. Those numbers describe a molecular method validated against CFU, and they show how CFU serves as the anchor currency even for methods that do not produce colonies.

Where CFU counting is strongest

Plate counts excel when the population is dense, the organism grows readily on agar, and the sample is clean enough to spread or filter. Heterotrophic plate counts of bottled water are a textbook case. In a survey of 21 Iranian bottled water brands sampled over three months, heterotrophic plate counts were measured in CFU/mL and averaged 16,855 ± 25,603 CFU/mL across the tested samples, a range that makes the point about variability in finished water [3].

Food growth studies rely on the same arithmetic. Naturally occurring Listeria in unpackaged retail ham reached a maximum of 1.6 × 10³ CFU/g, and growth of Listeria innocua in one sample proceeded at 0.02 log per hour to a ceiling of 4.0 to 5.0 × 10³ CFU/g [4]. Growth modeling of Pseudomonas aeruginosa in natural mineral water tracked populations to a maximum of 4 to 5 log CFU/mL, with the lower values attributed to microbial competition in the batch that started with a higher heterotrophic count [5]. In each case, the CFU value is the dependent variable, and the experiment is built around producing a countable plate.

Most Probable Number: A Statistical Estimate From Growth Patterns

Building the dilution series

An MPN experiment is a designed binary assay. You prepare a series of decimal dilutions of the sample, inoculate a fixed number of replicate tubes at each dilution (three and five are the most common configurations), and incubate them under conditions that let the target organism grow. Each tube is then scored positive or negative, usually by turbidity, gas production, a pH indicator, or a confirmatory transfer.

The scoring rule matters more than most people expect. A tube that is turbid because of a competitor is a false positive. A tube that stays clear because the target was injured and never resuscitated is a false negative. Both distort the MPN estimate in predictable directions.

Reading the MPN table

The pattern of positives is converted to a number using a table generated from the Poisson distribution and the assumption that organisms are randomly distributed in the sample. For a three-tube series, the pattern 3-2-0 (three positives at the highest concentration, two at the middle, none at the lowest) corresponds to a specific MPN and a specific 95% confidence interval.

Two features of these tables cause most of the confusion in practice.

First, the confidence interval is asymmetric. An MPN of 43 does not mean 43 ± 5. It means the true concentration plausibly lies within a range whose upper bound is farther from the estimate than its lower bound.

Second, the table is bounded. If every tube at every dilution is positive, the correct report is "greater than" the highest tabulated value, not a numeric estimate. If every tube is negative, the correct report is "less than" the lowest tabulated value. A well-designed series always brackets the true concentration so that the highest dilution is fully negative and the lowest is fully positive.

Enrichment-based MPN recovers stressed cells

The great practical advantage of MPN is that it can start with liquid enrichment, and liquid enrichment resuscitates. A cell that is too damaged to form a colony on a selective plate within 24 hours may still multiply in a non-selective or mildly selective broth over 24 to 48 hours and then declare itself.

The medium choice is not a minor detail. In a study of ready-to-eat seafood inoculated with low levels of Listeria monocytogenes, four selective enrichment media were compared, including half Fraser broth, two Listeria enrichment broth formulations, and buffered Listeria enrichment broth (BLEB). When samples were incubated in BLEB for 48 hours, MPN values agreed with the inoculation levels in 93 to 100% of cases across strains and food matrices, and MPN estimates for heat-injured inocula were accurate in 93% of cases [6]. The same dilution series read through a less supportive broth would have produced lower, wrong numbers. The lesson generalizes: in MPN, the enrichment medium is part of the measurement.

The comparison with plate counting is direct and well documented. In work on Listeria in ground meat, an MPN method based on the same media as the standardized plate count method had a detection limit about 66% lower than the plate method, which allowed detection of a clearly greater number of Listeria-positive samples from naturally contaminated ground meat. In the field trial on ground beef from Berlin retail stores, the MPN method yielded more Listeria spp.-positive samples (a relative 43% more) and more L. monocytogenes-positive samples (a relative 21% more) than the colony count method [7]. The same study found the standardized colony count method preferable for routine use because of slightly higher productivity and much smaller variation in results, while the MPN method was preferable for epidemiological work because of its lower detection level [7].

That is the trade in one paragraph. MPN buys sensitivity with precision.

Defined-substrate and miniaturized MPN

Modern MPN has moved well beyond racks of glass tubes.

Defined-substrate systems use a chromogenic or fluorogenic substrate that the target organism metabolizes to produce a detectable signal, with the MPN computed from the number of positive wells in a multi-well tray. In a seasonal study of Nile River water in Giza, the IDEXX Colilert-18 system gave a peak summer total coliform count of 21,462 MPN/100 mL, while traditional culture methods on the same samples gave 17,748 CFU/100 mL, with IDEXX consistently yielding 8 to 10% higher counts than the traditional methods [8]. That gap is not a defect in either method. It reflects differences in recovery, in the enzymatic definition of a positive, and in how the two methods handle sublethally injured cells.

Miniaturized MPN shrinks the format further. A model food developed to mimic plant-based meat analogues used a miniaturized MPN method to enumerate L. monocytogenes, chosen to minimize handling and disposable use while testing a large matrix of temperatures, water activities down to 0.940, and organic acid concentrations. The miniaturized MPN showed good correlation to CFU counting and allowed growth kinetics to be fitted across conditions that would have been impractical with full-scale tube racks [9].

Where MPN is strongest

MPN wins in four situations:

• The target is present at a concentration below the plating limit.
• The cells are injured, freeze-stressed, heat-stressed, or nutrient-starved.
• The matrix is too turbid, oily, or particulate for membrane filtration.
• The organism competes poorly with background flora on a solid selective medium.

Poultry rinse testing illustrates the first point precisely. In a study designed to identify poultry wing rinses carrying Salmonella at or above 10 CFU/mL, a three-tube MPN procedure had a lower limit of quantification of -0.96 log CFU/mL, and it overestimated the inoculated Salmonella level by only 0.05 ± 0.35 log CFU/mL across 120 rinses inoculated with cold-stressed and nutrient-starved strains at 0.22 to 3.79 log CFU/mL [10]. Two commercial rapid methods run on the same samples had higher quantification limits (1.00 log CFU/mL for GENE-UP QUANT and 0.00 log CFU/mL for BAX SalQuant), and one of them underestimated the inoculum by 1.21 ± 0.78 log CFU/mL [10]. MPN's accuracy at classifying rinses against the 10 CFU threshold was 93.1% [10].

Sub-1 CFU/mL quantification from a 3-tube series sounds impossible until you remember what MPN is doing. It is not counting organisms. It is estimating a rate from a Poisson model, and that model can resolve concentrations well below one per millilitre when the sample volume across all tubes is large.

The same principle appears in sprout testing. When alfalfa sprouts and spent irrigation water were collected during sprouting of naturally contaminated seed, Salmonella levels were determined by MPN analysis and background microflora by CFU, with background counts of 10⁷ to 10⁸ CFU/g or mL at 20 °C and 10⁸ to 10⁹ CFU/g or mL at 30 °C. The study authors then related detection performance of a PCR system and the FDA Bacteriological Analytical Manual method to those MPN levels, finding that above 1.1 MPN/g or mL both methods detected the organism in all samples, while at 0.21 MPN/g or mL or lower the PCR system detected more positive samples than the reference culture method [11]. MPN here serves as the quantitative backbone against which detection methods are judged.

When to Use CFU and When to Use MPN

Expected concentration and the lower limit of quantification

Start with the number you expect, not the number you hope for.

If the expected level is above roughly 10³ CFU per gram or mL, plating is the efficient choice. You get a direct number, a tight confidence interval, and a result in a day. If the expected level is below 10² per gram or mL, plating forces you to plate large volumes across many plates or to accept a "less than" report.

The seafood example is instructive. Because contamination levels of L. monocytogenes in ready-to-eat seafood are often below 10 CFU/g, accurate enumeration at low concentrations is essential for risk assessment, and MPN is the standard approach for that reason [6]. Attempting the same measurement by spread plating would require plating the entire sample and would still frequently return zero.

Sample matrix and competing flora

Membrane filtration fails on samples that clog the filter. Spread plates fail on samples that absorb into the agar unevenly or that carry particulates indistinguishable from colonies. MPN tubes do not care about turbidity.

Competing flora cuts the other way. On a solid selective medium, a fast-growing background population can overgrow a slow target. In liquid enrichment, the target gets a head start in a medium formulated to suppress competitors, and the subsequent dilution series dilutes the background along with the target.

Injured, stressed, and sublethally damaged cells

This is the single largest source of disagreement between the two methods in real samples. Food processing, chilling, drying, acid exposure, and chlorine treatment all produce cells that remain viable but lose the ability to form colonies on selective agar without a recovery step.

The ground beef Listeria comparison quantifies the effect. The MPN method's lower detection limit translated directly into more positive samples: a relative 43% more Listeria spp. positives and 21% more L. monocytogenes positives than the colony count method [7]. If your question is "is this organism present and roughly how much," MPN gives the more complete answer. If your question is "what is the concentration, with tight error bars," the plate method does [7].

Precision and agreement between laboratories

Plate counts are more reproducible. That is the consistent finding across comparative studies, and it is why standardized colony count methods remain the routine reference method in food microbiology even where MPN has a lower detection limit [7].

Interlaboratory proficiency testing makes the wider point. A food microbiology proficiency program with 450 participating laboratories assessed enumeration and detection of aerobic microorganisms, Enterobacteriaceae, coliforms, β-glucuronidase-positive E. coli, anaerobic sulfite-reducing bacteria, Clostridium perfringens, coagulase-positive staphylococci, and L. monocytogenes, and documented how method choice affects bacterial counts and their measurement uncertainty [12]. Two laboratories can analyze the same homogenate with different methods and produce legitimately different numbers. Reporting a count without the method is therefore incomplete, and comparing counts across methods without acknowledging that difference is a category error.

Turnaround time and throughput

Plate counts are fast for fast growers and slow for slow growers. A heterotrophic plate count reads in 24 to 48 hours. A Listeria plate count reads in 24 to 48 hours on a suitable medium.

MPN is slower at the low end because it depends on enrichment. The classic comparison makes this concrete: a three-tube MPN procedure for Listeria in fresh vegetables required 5 days, while an antibody-direct epifluorescent filter technique that correlated with both plate counts and MPN estimates over a range of 10 to 10⁷ CFU/mL could be completed in under 1 hour [13]. That study also noted that microbial backgrounds as high as 10⁸ CFU/mL did not affect the rapid method's performance, which is the kind of robustness claim an MPN procedure earns the hard way, through selective enrichment.

Reporting conventions and regulatory expectations

Conventions differ by sector, and following the local convention is not optional.

In drinking and recreational water microbiology, coliform results are conventionally reported as MPN per 100 mL, while heterotrophic plate counts are reported as CFU per mL. A survey of bottled water brands used MPN for total and fecal coliforms and heterotrophic plate counts for the general bacterial load in the same samples, reporting the two in their respective units without conversion [3]. A study of well designs in rural Tanzania reported total coliforms in CFU/100 mL and E. coli in MPN/100 mL within the same dataset, reflecting the different methods applied to each target [14]. A comparison of Enterococcus enumeration in marine beach samples reported defined-substrate and membrane filtration results jointly as "MPN or CFU/100 mL," which is the honest way to present data when the two methods are treated as equivalent for regulatory thresholds [15].

In food microbiology, CFU per gram is the default unit for enumeration, and MPN per gram appears where the target is expected at low levels or where the method is enrichment-based [11][6][9].

Worked Scenarios

Bottled water: HPC by CFU, coliforms by MPN

A finished bottled water sample needs two answers. The heterotrophic plate count asks how many total organisms will grow on a non-selective medium at a defined incubation temperature. The coliform screen asks whether indicator organisms are present at any level.

The first question is a plating question. The population may be anywhere from zero to tens of thousands per milliliter, and a plate count with a dilution series covers that range. In the Iranian bottled water survey, heterotrophic plate counts averaged 16,855 ± 25,603 CFU/mL with total and fecal coliforms negative by MPN across the tested samples, and the authors concluded that growth of microorganisms was possible in bottled water under favorable conditions [3].

The second question is an MPN question. Coliforms may be present at one organism per 100 mL, a level no practical plate count will detect. The MPN format, especially in a defined-substrate tray, handles that range directly.

Ground beef: routine monitoring versus epidemiological resolution

A routine ground beef Listeria screen wants a number that is reproducible between laboratories and comparable to historical data. The standardized colony count method delivers that, with smaller variation and slightly higher productivity than MPN [7].

An epidemiological investigation into contamination sources wants to know whether Listeria is present at all in samples that plate as zero. The MPN method detects a clearly greater number of positive samples from naturally contaminated ground meat because its detection limit is roughly 66% lower [7].

Same sample type, different question, different method. Neither choice is a compromise.

Poultry wing rinses: classifying against a threshold

Sometimes the required output is not a concentration but a classification. The FSIS Proposed Regulatory Framework to Reduce Salmonella Illnesses Attributable to Poultry highlighted the need for simple, rapid methods that identify poultry wing rinse samples harboring Salmonella at or above 10 CFU/mL, and a comparison of a three-tube MPN against rapid commercial methods and novel threshold tests was built around that exact requirement [10].

MPN quantified down to -0.96 log CFU/mL and was 93.1% accurate at classifying rinses against the threshold, but it is a multi-tube, multi-day workflow [10]. The threshold methods were developed specifically to answer the yes/no question faster. When the regulatory question is binary, the method should be binary too.

Ready-to-eat seafood: enumerating a pathogen below 10 CFU/g

A risk assessment for L. monocytogenes in ready-to-eat seafood needs a concentration, not a presence/absence call, and the concentration is often below 10 CFU/g [6]. That is an MPN problem by definition.

The medium selection work in that study shows how much accuracy depends on the enrichment broth: BLEB for 48 hours gave 93 to 100% agreement with inoculation levels and 93% accuracy on heat-injured inocula, outperforming the other three media evaluated for consistency across strains and matrices [6].

Surface water: comparing IDEXX MPN with plate counts

A monitoring program switching from traditional membrane filtration to a defined-substrate MPN system needs to know how the numbers will shift. In the Nile River comparison, the highest summer total coliform count was 21,462 MPN/100 mL by IDEXX and 17,748 CFU/100 mL by traditional methods, with IDEXX consistently 8 to 10% higher across the study [8]. Any historical trend line drawn across that method change has to account for the systematic offset.

Growth modeling in a plant-based matrix

A study of L. monocytogenes in a model plant-based meat analogue needed hundreds of enumerations across temperatures from 7 to 15 °C, water activities down to 0.940, and organic acid concentrations up to 2.5 mM lactic and 6 mM acetic acid. A miniaturized MPN method made that design feasible, correlated well with CFU counting, and allowed kinetic growth parameters to be determined across all matrices tested [9]. Here MPN is used not because the levels are low but because the format is compact.

Common Mistakes and Limitations

Treating MPN as a count. An MPN of 240 is not 240 organisms. It is a maximum-likelihood estimate derived from a tube pattern, and its confidence interval is wider and asymmetric. Reporting it as a count overstates precision and misleads anyone comparing it to plate data.

Converting MPN to CFU or the reverse. There is no valid conversion factor. The two quantities are defined differently, recover differently, and behave differently at the low end. When a method change requires comparability, generate a correlation curve on your own matrix. That is precisely how a validated alternative method is established under USP <1223>, where detection times from an automated growth-based system were plotted against log CFU values to establish numerical equivalence for yeasts and molds in personal care products [16].

Plating outside the countable range. A plate with 900 colonies is not a count of 900. It is a floor. A plate with two colonies is a count with a 70% coefficient of variation. Both should trigger a dilution change or an MPN.

Ignoring clumping. Reporting CFU as cells overstates the population when the organism grows in clusters or chains. This bias is systematic and one-directional.

Running an MPN series that does not bracket the concentration. If the lowest dilution is not fully positive or the highest is not fully negative, the table returns a censored value. All-positive and all-negative series are legitimate results, but they are not quantitative, and they should be reported as inequalities.

Choosing the wrong enrichment medium. The seafood work showed that the same MPN design read through four different selective broths produced different accuracy, with BLEB at 48 hours the only medium that consistently agreed with inoculation levels across strains and matrices [6]. An MPN result is only as good as the broth that produced the positives.

Ignoring viable but non-culturable cells. Neither CFU nor MPN recovers cells that have entered a viable but non-culturable state. Vibrio cholerae can enter that state under stress, and distinguishing VBNC cells from dead cells required a propidium monoazide treatment of 20 µM for 20 minutes combined with real-time PCR, optimized because the PMA effect varies among species and conditions [17]. A culture-based zero does not mean the organism is absent.

Assuming culture methods capture the true load. For slowly growing organisms, culture-based enumeration is both delayed and potentially biased low. A 16S rRNA-based molecular bacterial load assay adapted for mouse tissues produced results comparable to CFU counts in solid culture for Mycobacterium tuberculosis, while avoiding the delay imposed by the organism's slow growth and the risk of contamination that can destroy critical data points [18]. The point is not that molecular methods replace CFU. It is that CFU is one window on the population, not the population itself.

Reporting a number without its method and uncertainty. Proficiency testing across 450 laboratories documented how method choice affects bacterial counts and their measurement uncertainty [12]. A CFU/mL figure without the method, the medium, the incubation conditions, and the dilution scheme is not reproducible.

Comparing methods across a regulatory change without a bridging study. Large-scale comparisons of culture-based E. coli enumeration with a rapid qPCR method across more than 100 Michigan beach sites found correlation, with Pearson R² values of 0.452 to 0.641 across multi-site datasets, but correlation is not identity [19]. Similarly, qPCR and conventional Enterococcus testing in marine beach samples showed a positive correlation with an overall r of 0.71 while the reported units were different (gene copies versus MPN or CFU per 100 mL) [15]. Bridging studies are what make a method change defensible.

Limitations of scope. Individual sample results need interpretation in context, including the sampling plan, the analytical method, and the intended use of the data. When the sample comes from an animal, the count is only one input and individual case decisions belong with a veterinarian.

Frequently Asked Questions

Is MPN the same as CFU?

No. CFU is a direct count of colonies recovered on a plate, while MPN is a statistical estimate derived from the pattern of positive and negative tubes in a dilution series.

Can I convert MPN to CFU?

No reliable conversion factor exists. The two methods recover different fractions of the population and define a positive result differently. Comparability has to be established empirically on your own matrix and organism.

Which method is more accurate?

CFU counting is more precise and more reproducible, and it is preferred for routine enumeration. MPN is more sensitive at low concentrations and recovers stressed cells better, which makes it preferable for epidemiological and risk-assessment work.

Why is MPN reported per 100 mL for water?

Because coliform concentrations in treated water are frequently below one organism per 100 mL, a level that no practical plate count can resolve. The MPN format estimates concentrations in that range directly.

Can either method detect viable but non-culturable cells?

No. Both CFU and MPN depend on growth, so cells that have entered a viable but non-culturable state are missed by both. Viability PCR and molecular load assays address that gap.

Which method should a small laboratory set up first?

Set up plate counting first. It supports heterotrophic plate counts, food enumeration, and growth studies, and it is the reference against which alternative methods are validated. Add MPN when you need low-level detection or enrichment-based recovery.

If you are learning microbiology for the first time, the fastest way to internalize the CFU versus MPN distinction is to run both on the same sample and compare the reports side by side. The plate will give you a countable number with a tight range. The tubes will give you an estimate with a confidence interval. Once you have seen the two reports on the same bench sheet, the difference between cfu and mpn microbiology stops being an abstraction and becomes a reporting decision you can defend.

Campylobacter Enumeration: A Direct Method Comparison

Campylobacter enumeration is a useful case study because the organism is fastidious, sensitive to oxygen, and often present at levels where the choice of method changes the answer. A comparison of sampling techniques and enumeration methods for Campylobacter spp. in raw retail chicken legs examined 90 packages, rinsing one leg and homogenizing 25 g of skin from the other, and analyzed 40 of the 90 packages by both direct plating and MPN [20].

The prevalence differed by sampling technique: 70% (63 of 90) of skin samples and 77% (69 of 90) of rinse samples were Campylobacter-positive [20]. Direct plating produced a median of log 4 CFU per leg surface for skin samples with a standard deviation of 0.6, and a median of log 4.3 CFU per leg surface for rinse samples with a standard deviation of 0.9 [20]. Among the rinse samples with countable results, 73% (37 of 51) had higher numbers of Campylobacter than the paired skin samples, although the difference was not statistically significant (p = 0.08), and the correlation coefficient between skin and rinse counts was 0.43 [20]. The rinse samples were then used to compare direct plating against MPN, and in 5% (2 of 40) of the rinse samples the number of Campylobacter spp. could not be determined by direct plating at all [20].

That last figure is the practical argument for keeping both tools available. Direct plating produced quantitative data efficiently for most samples, but a subset fell outside the countable range and required the MPN format to be quantified. Compare quantitative data across chicken products are a major data gap for risk assessment, and the study was designed around closing that gap with methodologically sound numbers rather than convenient ones [20].

Reporting the Number Correctly

Several reporting habits separate a defensible result from a misleading one.

Report the unit with the method. CFU per gram and MPN per gram are different quantities. A data table that mixes them in a single column without a method flag creates a false sense of comparability.

Report the dilution scheme. For CFU, state the plated volume and the dilution. For MPN, state the number of tubes per dilution and the dilution series. A three-tube series and a five-tube series with the same positive pattern do not give the same estimate or the same confidence interval.

Report censored results as inequalities. "Less than the detection limit" and "greater than the highest tabulated value" are the correct forms. Substituting a number for these results fabricates precision.

Report the medium and the incubation conditions. The seafood enrichment comparison demonstrated that the same MPN design can produce different accuracy depending on the broth and the incubation time, and the difference was not subtle [6].

Report the confidence interval when the number is an MPN. This is not optional garnish. It is the measurement.

Putting the Decision Together

For most laboratories, the choice resolves into a short decision sequence.

Ask first what the expected concentration is. Above 10³ per gram or mL, plate. Below 10² per gram or mL, and especially below 10, run an MPN.

Ask second whether the cells are likely to be injured. Samples that have been heated, frozen, dried, acidified, or chlorinated carry a population that selective agar will undercount. Liquid enrichment recovers those cells, and the ground beef comparison quantified the payoff at a relative 43% more Listeria spp. positives and 21% more L. monocytogenes positives for MPN over the colony count method [7].

Ask third whether you need precision or sensitivity. Routine monitoring and trend analysis need reproducibility, which favors the plate method [7]. Source tracking, risk assessment, and regulatory threshold classification need sensitivity at low levels, which favors MPN [7][10][6].

Ask fourth what the reporting convention requires. Water microbiology expects MPN per 100 mL for coliforms and CFU per mL for heterotrophic plate counts [3][14]. Food microbiology expects CFU per gram for enumeration and MPN per gram for low-level pathogen work [11][6][9].

Ask fifth whether either method answers the actual question. If the organism may be viable but non-culturable, neither does, and a viability PCR or molecular load assay becomes part of the picture [18][17]. If the required output is a rate rather than a concentration, a growth model built on CFU or miniaturized MPN data may be the right frame [5][9].

The short version is that CFU tells you what grew and MPN tells you what was probably there. A laboratory that understands which of those questions it is asking will choose correctly almost every time.

Sources

1. Antibiotic resistance and growth of the emergent pathogen Escherichia albertii on raw ground beef stored under refrigeration, abuse, and physiological temperature.
2. Characterization of a three bacteria mixed culture in a chemostat: evaluation and application of a quantitative terminal-restriction fragment length polymorphism (T-RFLP) analysis for absolute and species specific cell enumeration.
3. Evaluation of chemical and microbiological quality in 21 brands of Iranian bottled drinking waters in 2012: a comparison study on label and real contents.
4. Enumeration and growth of naturally occurring Listeria spp. in unpackaged ham.
5. Growth kinetics modeling of Pseudomonas aeruginosa in natural mineral water.
6. Evaluation of selective enrichment media for accurate most probable number enumeration of Listeria monocytogenes in ready-to-eat seafood products.
7. [[Comparison of direct colony count methods and the MPN-method for quantitative detection of Listeria in model and field conditions].](https://pubmed.ncbi.nlm.nih.gov/11766274/)
8. Seasonal variations in microbial and physicochemical parameters of Nile river water in Giza, Egypt: comparison of IDEXX and conventional methods.
9. A model food mimicking plant-based meat analogues for determination of matrix effects on growth of Listeria monocytogenes.
10. Evaluation of Methods for Identifying Poultry Wing Rinses With Salmonella Concentrations Greater Than or Equal to 10 CFU/mL.
11. Evaluation of PCR detection of Salmonella in alfalfa sprouts and spent irrigation water collected during sprouting of naturally contaminated seed.
12. Lessons from the organization of a proficiency testing program in food microbiology by interlaboratory comparison: analytical methods in use, impact of methods on bacterial counts and measurement uncertainty of bacterial counts.
13. Comparison of antibody-direct epifluorescent filter technique with the most probable number procedure for rapid enumeration of Listeria in fresh vegetables.
14. Evaluation of Well Designs to Improve Access to Safe and Clean Water in Rural Tanzania.
15. Comparison of Enterococcus density estimates in marine beach and bay samples by real-time polymerase chain reaction, membrane filtration and defined substrate testing.
16. Efficacy Comparison of an Automated Growth-Based System and Plate-Count Method on the Detection of Yeasts and Molds in Personal Care Products.
17. Enumeration of viable non-culturable Vibrio cholerae using propidium monoazide combined with quantitative PCR.
18. Culture-Free Enumeration of Mycobacterium tuberculosis in Mouse Tissues Using the Molecular Bacterial Load Assay for Preclinical Drug Development.
19. Large-scale comparison of E. coli levels determined by culture and a qPCR method (EPA Draft Method C) in Michigan towards the implementation of rapid, multi-site beach testing.
20. Comparison of different sampling techniques and enumeration methods for the isolation and quantification of Campylobacter spp. in raw retail chicken legs.