We sampled forest floor herpetofaunal communities in a monsoonal rainforest in South India for three consecutive years to evaluate the use of cluster sampling in estimating species composition and density. Our initial experimental design consisted of comprehensive random searches of multiple 25m² quadrats (SRS) for animals. After our initial season we found that most quadrats had zero animals detected and, when encountered, animals were spatially aggregated. To increase sampling efficiency and derive more precise density estimates, we shifted to adaptive cluster sampling (ACS). We compared the relative sampling efficiencies of ACS to SRS and the ability of the 2 methods to detect rare species. Adaptive cluster sampling failed to yield the more precise density estimates as predicted by statistical theory. However, ACS yielded more individual and rare species detections. Our results suggest the ACS assumptions should be carefully evaluated prior to use because it may not be appropriate for all rare, spatially aggregated populations.