Chinese Chess is one of the most popular game in the world, computer Chinese Chess is important in computer artificial intelligence, computer evaluation function is a key factor for computer Chinese Chess, but the evaluation function is computed by a lot of parameters, each parameter is needed to be set a value, the traditional method is to tune values of evaluation parameters continuously by people, it is time-consuming, laborious, and difficult to be precise. This thesis researches the evaluation function for computer Chinese Chess, we design a new automatically tuning system of evaluation function by genetic algorithm and statistical method, it tunes values of evaluation parameters automatically, and divides into different stages by chess game. The experiment results show the system increases the power of our computer Chinese Chess program.