Abstract. Spatial heterogeneity, an important characteristic in semi-arid grassland vegetation, may be altered through grazing by large herbivores. We used Moran's I, a measure of autocorrelation, to test the effect of livestock grazing on the fine scale spatial heterogeneity of dominant plant species in the shortgrass steppe of northeastern Colorado. Autocorrelation in ungrazed plots was significantly higher than in grazed plots for the cover of the dominant species Bouteloua gracilis, litter cover and density of other bunchgrasses. No species had higher autocorrelation in grazed compared to ungrazed sites. B. gracilis cover was significantly auto-correlated in seven of eight 60-yr ungrazed exclosures, four of six 8-yr exclosures, and only three of eight grazed sites. Autocorrelograms showed that B. gracilis cover in ungrazed sites was frequently and positively spatially correlated at lag distances less than 5 m. B. gracilis cover was rarely autocorrelated at any sampled lag distance in grazed sites. The greater spatial heterogeneity in ungrazed sites appeared linked to patches characterized by uniformly low cover of B. gracilis and high cover of C3 grasses. This interpretation was supported by simple simulations that modified data from grazed sites by reducing the cover of B. gracilis in patches of ca. 8 m diameter and produced patterns quite similar to those observed in ungrazed sites. In the one exclosure where we intensively sampled soil texture, autocorrelation coefficients for sand content and B. gracilis cover were similar at lag distances up to 12 m. We suggest that the negative effect of sand content on B. gracilis generates spatial heterogeneity, but only in the absence of grazing. An additional source of heterogeneity in ungrazed sites may be the negative interaction between livestock exclusion and B. gracilis recovery following patchy disturbance.