In this paper, the concept of a tense tree and its construction method are proposed such that the issue of Chinese-English tense conversion is transformed into the issue of tagging a tense tree. Futher, tree-CRF is used to tag nodes of the untagged tense tree with English tenses. Templates of feature functions are suitable for the need of model inference. Experimental results show that the method of tree-based tense tagging is much better than linear-based tense tagging, and that tense trees can better express tense dependencies between clauses.