Although capture–recapture techniques are often used to estimate population size, these approaches are difficult to implement for a wide variety of species. Highly polymorphic microsatellite markers are useful in individual identification, and these ‘molecular tags’ can be collected without having to capture or trap the individual. However, several sources of error associated with molecular identification techniques, including failure to identify individuals with the same genotype for these markers as being different, and incorrect assignment of individual genotypes, could bias population estimates. Simulations of populations sampled for the purpose of estimating population size were used to assess the extent of these potential biases. Population estimates tended to be biased downward as the likelihood of individuals sharing the same genotype increased (as measured by the probability of identity (PI) of the multi-locus genotype); this bias increased with population size. Populations of 1000 individuals were underestimated by ≥5% when the PI was as small as 1.4 × 10−7. A similar-sized bias did not occur for populations of 50 individuals until the PI had increased to approximately 2.5 × 10−5. Errors in genotype assignment resulted in overestimates of population size; this problem increased with the number of samples and loci that were genotyped. Population estimates were often >200% the size of the simulated populations when the probability of making a genotyping error was 0.05/locus and 7–10 loci were used to identify individuals. This bias was substantially reduced by decreasing genotyping error rate to 0.005. If possible, only highly polymorphic loci that are critical for the identification of the individual should be used in molecular tagging, and considerable efforts should be made to minimize errors in genotype determination.