Recent studies of extra-pair paternity have found support for the idea that heterozygous males have an advantage in siring offspring. Most studies use DNA microsatellite loci to determine paternity and then use the same loci to estimate individual heterozygosity. However, because the likelihood of detecting extra-pair offspring depends on the combinations of parental alleles, it is possible that biases arise from particular allele combinations. This might produce false support for the influence of heterozygosity on mating behaviour. We used a simulation model to assess how large this bias might be. We found two sources of bias. First, we found a bias in the null hypothesis of a simple statistical test commonly used to test several predictions of the heterozygosity hypothesis. The use of randomization tests could eliminate this bias. Second, we found that using the same loci for both paternity and heterozygosity can cause an increase in results supporting the heterozygosity hypothesis when no effect of heterozygosity actually exists. This bias is reduced through the use of more markers with higher levels of polymorphism and heterozygosity, but can be eliminated entirely by using a separate set of markers to determine paternity and assess heterozygosity. The two sources of bias reduce evidence favouring the heterozygosity hypothesis, but do not negate all of the studies that support it. We suggest that further studies of heterozygosity and extra-pair paternity are important and likely to be informative, but our recommendations should be incorporated by researchers to improve the reliability of their conclusions.