The presence of autocorrelation invalidates all standard statistical tests unless special corrections are made. Because of this, it is important to know the degree of spatial autocorrelation in order to know how to sample. Mountain meadows were sampled to determine spatial autocorrelation of vegetation at the plant community level. A total of 40 meadows were sampled in the eastern Sierra Nevada, California. At each meadow a dominant plant community was selected for sampling. Sampling consisted of placing 10 × 10 cm quadrats at 1-m intervals on a 20-m transect and recording the presence for all vascular plant species rooted in the quadrats. Sites varied in plant species composition and number of species present. For each plot, ordination analysis in the form of reciprocal averaging was used to derive positions for each quadrat on axis 1. The scores from axis 1 were analyzed by semivariance to obtain the spatial dependence of the quadrats. Overall, three semivariance patterns were seen; A) plant communities that were autocorrelated at distances of less than one meter; B) communities that were autocorrelated between 1 m and 15 m; C) communities that were autocorrelated at distances greater than 20 m. Results indicate that for semivariogram type B, on average, sites were autocorrelated to a distance of 3.6 m, meaning that quadrats separated by greater than 3.6 m were independent. Beta diversity was significantly (P < 0.05) lower for semivariance type C than for either semivariance types A or B. These results are useful for determining spacing of sample points in mountain meadows to ensure spatial and statistical independence for presence/absence data.