The mark and recapture method estimates population size by catching a sample of animals, marking them, releasing them, and then catching a second sample later. The proportion of marked animals that show up in the second sample reveals how large the total population is. If you marked 100 fish and then caught 50 fish the next week, finding that 10 of them were marked, you’d reason that the marked fish are about one-fifth of whatever you catch, so the full population is probably around 500. That core logic has been the backbone of wildlife population estimates since the early twentieth century, but the method’s real-world reliability depends on a set of assumptions that are surprisingly easy to violate.
The Logic Behind the Estimate
The idea rests on a simple proportion. In the first visit, researchers capture and mark a known number of animals, then release them back into the population. After enough time has passed for the marked animals to mix back in with the unmarked ones, researchers return and capture a second sample. Some of those second-sample animals will carry marks from the first visit; the rest will be unmarked. The fraction of marked individuals in the second sample should roughly mirror the fraction of marked individuals in the entire population.
Turning that into a number is straightforward. You multiply the size of the first sample by the size of the second sample, then divide by the number of recaptured (marked) animals found in the second sample. This formula, known as the Lincoln-Petersen estimator, has been the standard starting point for capture-recapture work for over a century.1Biometrical Journal. The Petersen–Lincoln Estimator and its Extension to Estimate the Size of a Shared Population A small correction proposed by Chapman adjusts the result slightly to reduce bias when sample sizes are small, and that adjusted version is the one most field researchers actually use.2PLoS ONE. Reliability of Different Mark-Recapture Methods for Population Size Estimation Tested against Reference Population Sizes Constructed from Field Data
To make the fish example concrete: you tagged 100 fish (first sample). You came back and netted 50 fish (second sample). Of those 50, 10 carried tags (recaptures). Multiply 100 by 50, get 5,000, divide by 10, and you estimate 500 fish. If only 5 had been recaptured instead of 10, the estimate would double to 1,000. The fewer marked animals you recapture, the larger the implied population, which makes intuitive sense: your marked fish are a smaller drop in a bigger pond.
Assumptions That Have to Hold
The elegance of the method hides some demanding requirements. If any of them breaks down, the population estimate can drift far from reality, and the direction of the error depends on which assumption failed.
The first assumption is that the population is “closed” during the study. Between the marking visit and the recapture visit, no animals should enter (through birth or immigration) or leave (through death or emigration). In practice, this means the two visits need to happen close enough together that the roster of animals stays roughly the same, yet far enough apart that the marked animals have time to mix back in. For fast-breeding insects or highly mobile species, this window can be uncomfortably narrow.
The second assumption is equal catchability. Every individual in the population should have the same chance of being caught, whether it was caught before or not. This one is violated constantly. Some animals are bolder, some are shyer, and the experience of being caught can change behavior in either direction. A broad review of trap-dependence in vertebrate studies found that when animals do show a changed probability of recapture, “trap-happiness” — a higher chance of being caught again — is the rule rather than the exception, and it varies by sex, age, personality, and social status.3Oikos. Trap‐dependence in capture–recapture studies: empirical evidence in vertebrates and biological meaning If previously caught animals are more likely to be recaptured, the second sample will contain too many marked individuals, making the population look smaller than it really is. If they become trap-shy, the opposite happens.
The third assumption is that marks are not lost and are always detected. Animals lose ear tags, fin clips heal over, and paint fades. When a marked animal is recaptured but its mark has disappeared, it gets counted as a new, unmarked individual. That pushes the recapture count down and inflates the population estimate. In an elephant seal study, ignoring tag loss led to population estimates that were between 8% and 53% larger than corrected estimates, depending on how long the study ran and how quickly tags degraded.4PubMed Central. Complete tag loss in capture-recapture studies affects abundance estimates: An elephant seal case study
Finally, there is the mixing assumption. Marked animals need to distribute themselves randomly among unmarked ones before the second sample. If the marked animals clump together — say, because they were all caught at the same feeding site and returned to it — the second sample from a different location will contain few recaptures, and the population estimate will be artificially high.
When the Study Area Has Leaky Boundaries
One of the trickiest violations to handle is what researchers call “closure violation.” Many study areas have arbitrary boundaries: a forest plot, a stretch of river, a grid of hair snares. Animals do not know where the grid ends. Individuals near the edges drift in and out, which means the population being sampled is not really fixed.
A grizzly bear study using DNA-based mark-recapture illustrates how big the effect can be. Bears whose typical locations were within about 10 kilometers of the study grid edge showed lower fidelity to the study area and higher rates of new arrivals compared with bears living deep inside the grid. Naïve estimates that ignored edge effects were substantially higher than estimates based only on core animals extrapolated outward.5Canadian Journal of Zoology. Closure violation in DNA-based mark-recapture estimation of grizzly bear populations For wide-ranging species, this boundary problem is not a minor footnote — it can dominate the error budget.
Beyond Two Samples
The basic Lincoln-Petersen method uses exactly two sampling occasions. That works for quick snapshots, but many studies benefit from capturing animals across several sessions. The Schnabel census extends the logic to multiple capture events, accumulating information about marked and unmarked individuals over time. Each new session adds both newly marked animals and recaptures from all previous sessions, which tightens the estimate considerably.6Environmental and Ecological Statistics. Choice of sampling effort in a Schnabel census for accurate population size estimates Computer simulations have confirmed that when the equal-catchability assumption holds, the Schnabel method produces reliable estimates even for moderately-sized populations.7Ecological Modelling. Testing the effectiveness of capture mark recapture population estimation techniques using a computer simulation with known population size
The two-sample and Schnabel approaches both assume a closed population, though. For long-term monitoring where births, deaths, immigration, and emigration happen between sessions, researchers turn to open-population models. The Jolly-Seber family of models lets the population change between sampling occasions and separately estimates abundance, survival, and recruitment over time.8bioRxiv. Fast and flexible Bayesian Jolly Seber models and application to populations with transients These models require more data and make more statistical demands, but they reflect the reality that wild populations are not frozen in place.
Spatially Explicit Models
Traditional mark-recapture treats the study area as a featureless box: animals are either in it or not. In reality, animals have home ranges that overlap the study boundary to varying degrees, and an individual living mostly outside the trap grid might occasionally wander through and get caught once. Counting that animal the same way you’d count a resident introduces bias.
Spatially explicit capture-recapture models solve this by estimating where each individual’s activity center is, based on the pattern of which traps it was caught in. The result is a density estimate per unit area rather than a simple total count, which sidesteps the boundary problem almost entirely.9Ecosphere. Understanding spatially explicit capture–recapture parameters for informing invasive animal management These models have become especially popular for large carnivores. Tiger and leopard studies, for instance, now routinely fit spatially explicit models to camera-trap data, incorporating habitat covariates to produce density maps that inform conservation planning.10PubMed Central. Density estimation of tiger and leopard using spatially explicit capture-recapture framework
Replacing Physical Tags With DNA and Photographs
Physically catching and marking animals is expensive, stressful for the animals, and sometimes impossible. Over the past two decades, “non-invasive” mark-recapture has exploded. Instead of putting a tag on an animal, researchers collect something the animal leaves behind or photograph a feature that makes it individually recognizable.
DNA-based studies typically use hair snares, scat collection, or environmental sampling. An animal that rubs against a barbed-wire snare leaves behind hair follicles; genetic analysis identifies the individual, effectively “marking” it without anyone ever handling it. The encounter histories that result are analyzed the same way as traditional capture data, but hierarchical models account for the added uncertainty from genotyping errors.11PubMed. Hierarchical models for estimating density from DNA mark-recapture studies The grizzly bear study mentioned earlier was one of the pioneering examples of this approach.
Photo-identification takes advantage of the fact that many species carry unique natural markings: the spot patterns of whale sharks, the facial ridges of amphibians, the stripe arrangements of tigers. Researchers photograph individuals and then match new photos against a library to determine which animal is which. Software tools have been developed to automate the matching process, using image-recognition algorithms that extract scale-invariant features from each photo and rank potential matches.12Methods in Ecology and Evolution. A computer‐assisted system for photographic mark–recapture analysis One system applied to whale sharks used an information-theoretic approach to rank candidate matches, allowing researchers to maintain a growing library of individuals without handling the animals at all.13PubMed Central. Spot the match – wildlife photo-identification using information theory Other software has been specifically designed to minimize the amount of manual photo preprocessing so that researchers can scale up the approach to larger populations more quickly.14Ecological Informatics. APHIS: A new software for photo-matching in ecological studies
Citizen science is beginning to enter this space, too. A study on amphibian photo-identification asked anonymous volunteers to match facial images of individual animals. Participants averaged 26 out of 30 correct matches, and most completed the task in under a minute per image, suggesting that crowdsourced photo-matching could supplement automated systems for species where algorithms struggle.15PubMed Central. Using citizen science in the photo-identification of adult individuals of an amphibian based on two facial skin features
Practical Concerns in the Field
Before anyone sets a trap, one of the most important decisions is how many animals to mark and how many to recapture. Sample sizes that are too small produce estimates with confidence intervals so wide they are practically useless. Classic fisheries guidance showed that for a single mark-recapture experiment to achieve a desired level of precision, the product of the two sample sizes needs to exceed a threshold related to the population size, and the sampling effort should ideally be split evenly in cost between the marking and recapture phases.16Transactions of the American Fisheries Society. Sample Size in Petersen Mark–Recapture Experiments Follow-up work developed simpler equations that field biologists could apply without extensive computation, making it more practical to plan sample sizes before heading into the field.17Transactions of the American Fisheries Society. Sample Sizes for Single Mark and Single Recapture Experiments The catch is that you need at least a rough guess at the population size to plan your sample sizes, which creates a circular problem that usually gets resolved with pilot data or expert judgment.
The marking technique itself matters more than many people realize. For small fish, visible implant elastomer (VIE) tags — tiny injections of colored, flexible material under the skin — are a common choice. Researchers have refined the technique to include analgesic compounds during tagging and antiseptic aftercare to improve healing and survival.18PubMed Central. Identification of Individual Zebrafish (Danio rerio): A Refined Protocol for VIE Tagging Whilst Considering Animal Welfare and the Principles of the 3Rs For mammals and birds, external devices like collars, bands, or GPS tags are common, but the physical burden on the animal is a genuine concern. A study measuring the forces exerted by tags on carnivores during movement found that tags nominally set at the widely used 3% body-mass guideline produced forces equivalent to 4–19% of the carrier’s body mass while the animal was moving, reaching as high as 54% in a sprinting cheetah.19PubMed Central. Animal lifestyle affects acceptable mass limits for attached tags If tags change how an animal behaves, feeds, or evades predators, they violate the equal-catchability assumption and the assumption that marking does not affect survival.
Counting People Instead of Animals
The same mathematical logic works for estimating the size of human populations that are hard to count directly, such as people living with an undiagnosed disease, undocumented immigrants, or individuals engaged in illegal activity. Instead of physically marking people, researchers use overlapping administrative lists. Each list functions as a “capture occasion.” A person who appears on two or more lists is a “recapture,” and the proportion of overlap between lists lets you estimate how many people are not on any list at all.20PubMed Central. Evaluating tools for capture-recapture model selection to estimate the size of hidden populations
A South Korean study used three registries — health insurance claims, a national HIV/AIDS notification system, and a cohort study database — to estimate the total number of HIV-positive individuals in the country. The registries collectively identified about 2,300 people, but the capture-recapture analysis estimated roughly 14,900 additional HIV-positive individuals who did not appear on any of the three lists, putting the real total at around 17,100 and the combined completeness of all three registries at only about 13%.21PubMed Central. Using the capture-recapture method to estimate the human immunodeficiency virus-positive population That kind of finding has direct policy implications: it tells public health agencies how much of the burden they are missing and where to target testing efforts.
People have tried applying capture-recapture logic to software testing, too — treating each testing team as a “capture occasion” and each bug found by multiple teams as a “recapture” — to estimate how many bugs remain undiscovered. The results have been mixed. One critical analysis concluded that the method tends to overestimate the number of bugs to be found and breaks down especially in the later stages of testing, when remaining bugs are rare and hard to seed convincingly.22ScienceDirect (Journal of Systems and Software). A criticism on the capture-and-recapture method for software reliability assurance The assumptions translate awkwardly from ecology to code: bugs are not identical in difficulty, testing teams are not independent in their approach, and the “population” of bugs changes as fixes are applied. The method works better for counting hidden people than hidden software defects.
Where the Method Struggles Most
Mark-recapture is weakest for species that are extremely rare, extremely mobile, or extremely trap-averse. When recapture rates are very low, the denominator of the basic formula becomes tiny, and small changes in the number of recaptures swing the estimate wildly. Imagine catching 200 animals and marking them, then catching 200 more and finding just 1 recapture versus 2 recaptures. The estimate jumps from 40,000 to 20,000 based on a single animal. For populations where individual encounters are that scarce, confidence intervals become enormous and the estimate offers little practical guidance.
Highly social species pose their own problems. If animals move in groups, capturing one means you tend to capture its associates, violating the assumption that captures are independent. Colonial-nesting birds, schooling fish, and pack-living mammals all introduce this kind of correlation. Researchers sometimes handle it by treating the group as the sampling unit rather than the individual, but that requires additional information about group size and composition.
Despite these limitations, mark-recapture remains one of the only tools available for estimating the total size of populations that cannot be counted directly. Census methods — literally counting every individual — work for species that gather in visible aggregations, like penguin colonies or roosting bats, but most wild populations are too dispersed and secretive for that. Distance sampling, which estimates density from how far away detected animals are, offers an alternative for some species, but it has its own demanding assumptions about detection probability. Mark-recapture’s enduring appeal is that it turns a solvable proportion problem into a population estimate, and its century of refinement means there is now a model variant available for almost every complication the field can throw at it.