<p><em>Face recognition is a process of identification with the image has variations changeable can be recognized, needs a method of optimization to minimize computational time by not affecting the classification results</em><em>. This research proposes a face recognition system are directly based on Kohonen SOM classification that optimized by the method of Discriminant Analysis based Principal Component Analysis (PCA). Evaluation of PCA’s extraction performance uses two approaches, first the LDA method to optimize PCA issues of the election of irrelevant features of the dataset and the second approach is to apply a kernel function on the LDA (KDA), the results of both approaches are applied on face image classification for Kohonen directly. The testing is two phases, the first stage is testing with a single image of a face and then multi face. Based on the results of testing one face image, both of the approached feature extraction that proposed is very accurately be applied to the classification of the Kohonen SOM with the accurate value of the second approach PCA-KDA is more accurate with 94.22% and the first approach 93.91%, however on the first approach is faster than the second approach with the accurate value of time 0.4 seconds for PCA-LDA and 0.5 seconds to PCA-KDA to one image of the face, but while testing of multi face more two images the result is not significant.</em></p><p><em> </em><strong><em>Keywords:</em></strong> <em>Face recognition, Feature extraction, Kohonen SOM.</em></p>