It is clear that facial expressions are important on face-to-face communication. Humans can not only guess other person's psychological state from facial expressions but can also give own intentions to other persons by using them. If users can use facial expressions to a computer and the computer can correctly recognize them, an effective human interface can be constructed. Therefore, it is necessary to establish an effective facial expression recognition technique from facial image data. In this paper, we propose an application of the MTS (Mahalanobis-Taguchi-System) method for facial expression recognition. Especially, we treat the facial expression recognition from stationary facial image data. The MTS method is a statistical pattern recognition method using the Mahalanobis distance. This method can consider the correlation among each attribute in the recognition, and can optimize the number of attributes based on the results of recognition. We also propose an optimization method with the genetic algorithms. We construct the recognition system based on the MTS method, and then evaluate the effectiveness of the proposed methods through practical experiments. The experimental results revealed that our proposed methods could correctly recognize the facial expressions and could effectively optimize the number of attributes.