Comparisons of species richness among assemblages using different sample sizes may produce erroneous conclusions due to the strong positive relationship between richness and sample size. A current way of handling the problem is to standardize sample sizes to the size of the smallest sample in the study. A major criticism about this approach is the loss of information contained in the larger samples. A potential way of solving the problem is to apply extrapolation techniques to smaller samples, and produce an estimated species richness expected to occur if sample size were increased to the same size of the largest sample. We evaluated the reliability of 11 potential extrapolation methods over a range of different data sets and magnitudes of extrapolation. The basic approach adopted in the evaluation process was a comparison between the observed richness in a sample and the estimated richness produced by estimators using a sub-sample of the same sample. The Log-Series estimator was the most robust for the range of data sets and sub-sample sizes used, followed closely by Negative Binomial, SO-J1, Logarithmic, Stout and Vandermeer, and Weibull estimators. When applied to a set of independently replicated samples from a species-rich assemblage, 95% confidence intervals of estimates produced by the six best evaluated methods were comparable to those of observed richness in the samples. Performance of estimators tended to be better for species-rich data sets rather than for those which contained few species. Good estimates were found when extrapolating up to 1.8-2.0 times the size of the sample. We suggest that the use of the best evaluated methods within the range of indicated conditions provides a safe solution to the problem of losing information when standardizing different sample sizes to the size of the smallest sample.