This paper presents a forest inventory study of the Mao-shan area, a region which is partly representative of the forest types of southern China. In the study, the effectiveness of various feature extraction techniques was investigated, new classification algorithms were developed and supervised classification schemes were implemented and assessed. A scheme involving two-dimensional spectrum decomposition classification, manual editing and Bayes classification is proposed. Its application gives results which show promising potential for forest inventory in southern China using computer processing of LANDSAT imagery.