Abstract: A character choosing method is proposed based on the Boosting algorithm. The effect of the character (EC) depends on the train error, convergence speed and the test error. The character can be ordered by the effect of the character, and the redundancy character is deleted. Character choosing can reduce the dimension of the character space. The residual character is weighted by the effect of the character, and the exactness of object recognition is improved. The method will not lead to overfit like in the case of other classification learning machines, and the model of the classification is small, and is suitable for object recognition where the character is not obviously determined.