This paper explores the problem of parsing Chinese long sentences. Inspired by human sentence processing, a second-stage parsing method, referred as main structure parsing in this paper, are proposed to improve the pars-ing performance as well as maintaining its high accuracy and efficiency on Chinese long sentences. Three different methods have at-tempted in this paper and the result shows that the best performance comes from the method using Chinese comma as the bounda-ry of the sub- sentence. According to our ex-periment about testing on the Chinese de-pendency Treebank 1.0 data, it improves long dependency accuracy by around 6.0 % than the baseline parser and 3.2 % than the previ-ous best model. 1