Abstract: Decision tree is an optimization method. By simulating the nature evolution, genetic algorithm becomes a kind of general purpose global search algorithm. An approach to construct and optimize decision tree based on genetic algorithm is discussed in the paper here. A genetic algorithm is employed at each nonterminal node of the binary tree to search for weight vector. To achieve high performance, its fitness function and encoding method are discussed in detail, and in particular, much emphasis is put on its operator evolution. The effects of the misclassification rate of the binary tree are discussed and the experiment results show that the binary decision tree constructed based on genetic algorithm can classify sample set accurately.