By using moving split-window techniques (MSWT), this study estimated how far the edge effects penetrated the forest and agricultural fields in the arid valley of upper reaches of Minjiang River, southwestern China. Its aim was to provide general information on vegetation along edge to interior gradients in order to assist in interpretation and prediction of biological phenomena associated with agriculture-forestry boundary, and to improve current management practices in such areas. Three types of boundaries (10 transects) were investigated and sampled. The results showed that when the window width reached 6-10, the change of the SED curve on the graph tended to become stable, and one or two peaks occurred. The depth of edge influence was clearly different for different types of boundaries, and could be estimated within 50 m from the edge to interior. The depth of edge influence (DEI) on vegetation diversity almost varied between 12-30 m, mainly depending on the patch type, topography and microclimate, but seldom on slope orientation. Of the 6 forest transects in the three types of boundaries, the DEI was detected only in the forest part transects M2 and M6, but almost detectable in the agricultural part of all transects. MSWT was considered to be a useful tool for characterizing edge dynamics if enough data was available, and became a simple and powerful technique for analyzing the boundary. The results will provide further knowledge for understanding the interaction between forestry and agriculture in the arid valley.