The current rate of species attrition necessitates the development of quick and accurate sampling protocols and species richness estimators. Four time-based and one area-based methods were used to sample spiders of a grass bald and a heath bald in the Great Smoky Mountains National Park in late spring and early fall of 1995. Eighty-four samples were collected at each site; 1853 adults and 91 species were found in the grass bald, 573 adults and 60 species in the heath bald. The data were analyzed with 11 species richness estimators: Chao & Lee 1, Chao & Lee 2, ACE, ICE, bootstrap, Chao 1, Chao 2, first-order jackknife, second-order jackknife, Michaelis-Menten runs, and Michaelis-Menten means. All but the Chao & Lee estimators generated richness estimates that clustered within a reasonable range, 106–160 species for the grass bald and 68–90 species for the heath bald. The failure of the observed species accumulation curve to level off for our data sets showed that more sampling would be needed to determine the number of species present as adults during the two sampling seasons. Although this prevented us from rigorously testing richness estimator performance, we found that the Michaelis-Menten means estimator performed better than the other estimators when judged by two indirect criteria of good estimator performance—the estimator curve should reach an asymptote with fewer samples than are required for the observed species accumulation curve to reach an asymptote, and the estimates should be close to reasonable visual extrapolations of the asymptote of the observed species accumulation curve. We postulate that the differences we found in species richness and taxon and guild composition between the spider assemblages of these two bald communities are, at least in part, a consequence of striking differences in the physiognomy, richness, and taxonomic composition of the plant associations of the two communities.