Highly informative genetic markers, such as simple sequence repeats (SSRs), can be used to directly measure pollen flow by parentage analysis. However, mistyping (i.e. false inference of genotypes caused by the occurrence of null alleles, mutations, and detection errors) can lead to substantial biases in the estimates obtained. Using computer simulations, we evaluated a direct method for estimating pollen immigration using SSR markers and a paternity exclusion approach. This method accounts for mistyping and does not rely on assumptions about the distribution of male reproductive success. If ignored, even minor rates of mistyping (1.5%) resulted in overestimating pollen immigration by up to 150%. When we required at least two mismatching loci before excluding candidate fathers from paternity, the resulting pollen immigration estimates had small biases for rates of mistyping up to 4.5%. Requiring at least three mismatches for exclusion was needed to minimize the upward biases of pollen immigration caused by rates of mistyping up to 10.5%. The minimum number of highly variable SSR loci needed to minimize cryptic gene flow and obtain reliable estimates of pollen immigration varied from five to seven for a sampling scheme applicable to most conifers (i.e. when paternal haplotypes can be unambiguously determined). Between five and nine highly variable SSR loci were needed for a more general sampling scheme that is applicable to all diploid seed plants. With moderately variable SSR markers, consistently accurate estimates of pollen immigration could be obtained only for rates of mistyping up to 4.5%. We developed the pollen flow (pfl) computer program which can be used to obtain unbiased and precise estimates of pollen immigration under a wide range of conditions, including population sizes as large as 600 parents and mistyping rates as high as 10.5%.