Parentage analysis in natural populations presents a valuable yet unique challenge because of large numbers of pairwise comparisons, marker set limitations and few sampled true parent–offspring pairs. These limitations can result in the incorrect assignment of false parent–offspring pairs that share alleles across multi-locus genotypes by chance alone. I first define a probability, Pr(δ), to estimate the expected number of false parent–offspring pairs within a data set. This probability can be used to determine whether one can accept all putative parent–offspring pairs with strict exclusion. I next define the probability Pr(φ|λ), which employs Bayes’ theorem to determine the probability of a putative parent–offspring pair being false given the frequencies of shared alleles. This probability can be used to separate true parent–offspring pairs from false pairs that occur by chance when a data set lacks sufficient numbers of loci to accept all putative parent–offspring pairs. Finally, I propose a method to quantitatively determine how many loci to let mismatch for study-specific error rates and demonstrate that few data sets should need to allow more than two loci to mismatch. I test all theoretical predictions with simulated data and find that, first, Pr(δ) and Pr(φ|λ) have very low bias, and second, that power increases with lower sample sizes, uniform allele frequency distributions, and higher numbers of loci and alleles per locus. Comparisons of Pr(φ|λ) to strict exclusion and CERVUS demonstrate that this method may be most appropriate for large natural populations when supplemental data (e.g. genealogies, candidate parents) are absent.