Spatial pattern heterogeneity exhibits various patterns at different scales; therefore different observed scales affect the results of spatial analysis. Effects of changing scale on spatial analysis have been studied for decades in geography and ecology, but few previous studies expounded the quantitative approach about the optimum grain. Some studies describe the eroded landscape pattern neglecting the grain effects, which may lead to the difference deviation. The main goal of this study was to set the optimum grain and study the pattern change under this grain, and the dynamic features of eroded landscape pattern in Zhuxi small watershed were researched by 3S and FRAGSTATS3.3. In order to get the optimum grain, 14 metrics were calculated in 19 different multiple grain sizes, and 10 m is finally defined as the optimum grain by characteristic curves of landscape metrics corresponding to grain sizes. Based on the optimum grain, a series of metrics in both class level and landscape level were chosen to describe the pattern change of Zhuxi small watershed. During 1999-2007, erosion types with higher intensity were largely transformed into micro-degree erosion and light erosion; small eroded patches of micro-degree erosion and light erosion had expanded to form large continuous ones with less fragmented, more regular and concentrated; patches of moderate, intensive, very intensive erosion types were becoming smaller, isolated and disconnected; the figures were becoming more regular and less complicated. At the landscape level, SHDI, SHEI, SHAPE_MN, FRAC_MN and AI decreased while PD, ED and CONTAG increased, which leaded to decreased landscape heterogeneity, more discrete and complex spatial distribution of patches, more regular shape and landscape fragmentation increase.