In this paper, we propose a novel selftraining strategy for parsing which is based on Treebank conversion (SSPTC). In SSPTC, we make full use of the strong points of Treebank conversion and self-training, and offset their weaknesses with each other. To provide good parse selection strategies which are needed in self-training, we score the automatically generated parse trees with parse trees in source Treebank as a reference. To maintain the constituency between source Treebank and conversion Treebank which is needed in Treebank conversion, we get the conversion trees with the help of self-training. In our experiments, SSPTC strategy is utilized to parse Tsinghua Chinese Treebank with the help of Penn Chinese Treebank. The results significantly outperform the baseline parser. 1