Despite the importance of the soil seed bank in tropical forest regeneration, little is known about spatial variability in species composition and abundance of seeds stored in the soil. To develop sampling methods for comparative studies, we examined species richness, spatial variation, and abundance of germinants from the soil seed bank in a 16 year old secondary, tropical wet forest at La Selva Biological Station, Costa Rica. Surface soil (10 cm deep, 4.7 cm diameter) was collected at the intersection points of a gridded 1 ha plot (10 × 10-m grid, 121 samples) and in a nested 100 m2 subplot (2 × 2-m grid, 36 samples). The 1 ha plot had a density of 4535 seeds/m2 with 34 species observed. Based on a series of 100 randomized species accumulation curves, a Michaelis-Menten fit predicted a mean species richness of 36.3 species; the number of observed species was close to the predicted asymptote. A nonparametric, first-order jackknife species richness estimtor predicted a species richness of 37.0 species. Eighty-five and 95 percent of the observed species richness is contained, on average, within 41 and 74 pooled samples, respectively. Within the 100 m2 nested subplot, a density of 5476 seeds/m2 was observed, comprising 26 species with an estimates species richness (Michaelis-Menten fit) of 29.1 species. The jackknife species richness estimator predicted a species richness of 36.7 species. For species richness and abundance of both plots, spatial autocorrelation statistics (Moran's I) were not significantly different from zero at lag distances from 2 to 100 m, indicating a random distribution at these spatial scales. For this site, accurate estimates of species composition depend upon the number of samples collected as well as the spatial distribution of sampling effort. Many small samples distributed over a large area provide greater accuracy and precision for estimating species richness of the soil seed bank.