Forest stands classification is the precondition of stand inventory and forest management. It is also the primary task and main content of the inventory and dynamic monitoring of forest resources. With the improvements of the spatial resolution of remotely sensed image and the development of GEOgraphic Object-Based Image Analysis (GEOBIA) technologies, the precision of stand division had increased. However, the current methods can only classify forest stand coarsely, and they are difficult to classify the stands with different spatial structures. In this study, three indices describing multi-scale structure features, trees crown coverage (TCC), trees cluster index (TCI) and ratio of shadows and trees (RST), were derived from high spatial resolution image by GEOBIA technology. The indices can well indicate the diversities of stand density, structure, size and differences in the spatial distribution. With Quickbird image, we tested the method in Tekesi forest farm of Xinjiang province, P. R. China. The result showed that the stands with the difference spatial distribution could be classified according to these indices. Using the method to extract stands, can improve efficiency, reduce costs, and keep consistency of the results, which are essential for dynamic monitoring and changing detection of forest resources. The methods of indices extraction and classification are also suitable for similar applications that need the structural features accounting for the density and aggregation of the objects and their shadows, such as the urban area division.