Many techniques have been developed to estimate species richness and beta diversity. Those techniques, dependent on sampling, require abundance or presence/absence data. Palaeontological data is by nature incomplete, and presence/absence data is often the only type of data that can be used to provide an estimate of ancient biodiversity. We used a simulation approach to investigate the behaviour of commonly used similarity indices, and the reliability of classifications derived from these indices, when working with incomplete data. We drew samples, of varying number and richness, from artificial species lists, which represented original life assemblages, and calculated error rates for classifications of the parent lists and samples. Using these results, we estimated the Minimum Sample Richness (MSR) needed to achieve 95% classification accuracy. Results were compared for classifications derived from several commonly used similarity indexes (Dice, Jaccard, Simpson and Raup-Crick). MSR was similar for the Dice, Jaccard and Simpson indices. MSR for the Raup-Crick index was often much lower, suggesting that it is preferable for classifying patchy data, however the performance of this index was less stable than the other three in the simulations, which required an even lower MSR. MSR can be found for all presence/absence data from the contour graphs and equations as long as the absolute species richness and the beta diversity can be estimated.