Abstract: Rule-base optimization is very important for improving the translation quality of a transfer-based MT system. In this study, a new metric for rule evaluation is proposed and the strategy of adding rules incrementally in the process of optimization is applied to decrease the effect of crossing-constraints. Experimental results show that the assessment score of open corpus improves by 3.699% under 5-gram Nist metric when the new optimization method is used. And the optimization performance is better than other methods.