Due to greedy algorithm is adopted in the analysis of dependency parsing based on transformation, thus we select the most likely analysis action in each analysis step and ignore the dependency relation analysis of sentences. Therefore, in this paper, we propose decision based multi-candidate Chinese dependency syntax method on the basis of traditional Are-eager algorithms proposed by Nivre. In the new method, multiple candidate structures are introduced. We select multiple possible motion analysis according to every analysis step, and the optimal solution is selected in the sentence, in order to determine the dependency relation between different words and improve the correct rate. We choose the Chinese Pennsylvania Treebank and establish dependency analyzer based on support vector machine, we prove the effectiveness of the algorithm.