A combined method for automatic face recognition with both independent component analysis (ICA) and genetic algorithms (GA) was proposed. The fourth order blind identification (FOBI) algorithm was used to derive the independent sources out of the face images. To decrease the computing complexity, the dimension of the original image was reduced and GA was applied to the sources set to get an optimal subset of it. Faces can be recognized by classifying the coefficients of the face image projecting to the independent bases. The experiments in the face database show both a higher recognition rate than the eigenface method based on PCA and a lower computing complexity than the traditional ICA face recognition method.