ARISA (automated ribosomal intergenic spacer analysis) is a commonly used method for microbial community analysis that provides estimates of microbial richness and diversity. Here we investigated the potential biases of ARISA in richness estimation by performing computer simulations using 722 complete genomes. Our simulations based on in silico PCR demonstrated that over 8% of bacterial strains represented by complete genomes will never yield a PCR fragment using ARISA primers, usually because their ribosomal RNA genes are not organized in an operon. Despite the tendency of ARISA to overestimate species richness, a strong linear correlation exists between the observed number of fragments, even after binning, and the actual number of species in the sample. This linearity is fairly robust to the taxon sampling in the database as it is also observed on subsets of the 722 genome database using a jackknife approach. However, this linearity disappears when the species richness is high and binned fragment lengths gradually become saturated. We suggest that for ARISA-based richness estimates, where the number of binned lengths observed ranges between 10 and 116, a correction should be used in order to obtain more accurate "species richness" results comparable to 16S rRNA clone-library data.