The purpose of this study is to propose a criterion for determining the number of common factors. This criterion is called MSCV (Model Selector by Cross Validation), which is an estimator of expected log-likelihood. Through the results of computer simulations, the criterion is shown to be effective for determining the number of common factors in factor analysis. The practical application with such criterion is demonstrated by using standardized data of Japanese WISC-R, made of three common factors.