Abstract: The present implement of remote sensing in practical forest inventory dont take advantage of imagery processing and analysis technology of remote sensing, and is inefficient. Aimed to the auto-delineation of the boundary of forest sub-compartment for actual application, an imagery segmentation method was studied. SPOT5 imagery was segmented by four protocols, evaluation with nine features separately based on ultimate measurement accuracy, and synthetical evaluation of similarity based on Euclidean distance(ED) were used to evaluate the segmentation, here the features were roundness(RO), compactness(CO), shape index(SI), radius of smallest enclosing ellipse(RE), elliptic fit(EF) and form factor(P2A), relative error of area(RA), relative error of perimeter(RP) and displacement of the center of polygon(DC), while area(A), perimeter(P), RO, CO, SI, RE, EF, P2A were used to calculating ED. The result indicated that segmentation on raw imagery was better than that on histogram equalization imagery, the weights of input layers would affect the output of segmentation, and it would get better result by segmentation with the weight based on the standard deviation than that based on the information content of input imagery layers. It is feasible that preparing draft map for forest resources inventory through boundary auto-delineation based on imagery segmentation, for it is not only efficient and low labor intensity, but also improve the division precision of boundary and area accuracy of forest sub-compartment.