The partition ambiguity of translation derivations is an important problem suffered by the statistical machine translation, and it is much more important in a hierarchical phrase-based machine translation. In the paper, a hierarchical partition model is proposed to address the problem. The study applies markov random fields to construct the model, and integrate it into the hierarchical translation model to automatically select the more reasonable partition. In the NIST Chinese-English translation tasks, the optimization of the model is very efficient, and it improves the translation performance for hierarchical phrase-based translation on NIST05, NIST06 and NIST08 test sets.