Face recognition technology is attracting attention as one of the next generation biometrics technologies. Face recognition technology does not need a precise recognition device and recognizes the object by non-contact using a camera. Therefore, the object is attracting attention due to the advantage that it is less discomfort and inconvenience, and active research is proceeding. However, the conventional face recognition technology that performs face recognition using a fixed camera is less reliable in correct face recognition due to false detection due to a change in pose (rotation change) of the subject. Therefore, it is necessary to learn pose change for designing a face recognition system that is robust to pose change. In this paper, we propose a method to perform face recognition by using pose estimation and K ? Nearest Neighbor (KNN) to compensate for the disadvantages of face recognition for such pose change. KNN is advantageous for pose estimation through comparison with all data. However, because the amount of calculation increases as the number of data increases, the processing speed of pose classification is improved by reducing the dimension of image through principal component analysis. The pattern classification algorithm for face recognition uses a soft max pattern classifier based on FCM-RBFNNs. The structure of the FCM-RBFNNs-based soft-max pattern classifier is the same as that of the conventional FCM-RBFNNs, but has a difference in the learning method. Unlike the conventional RBFNN, the proposed pattern classifier uses the learning method of the logistic regression model. (2D) 2PCA is used for data dimension reduction to perform face recognition. In order to evaluate the performance and pose classification performance of the proposed FCM-RBFNNs-based soft-max pattern classifier, we used the Cambridge Head Pose Image Database, CMU Pose Illumination Expression Database, and Yale Dace Database B.