Analysis of medical images is done by using image processing as it is one of the prominent tools. It is also used in the breast cancer detection. This experiment focuses towards the identification of relevant, representative and more important, discriminate image features for analysis of medical images. The features from mammogram images representing normal tissues or benign and malign tumors are extracted using Gabor wavelets. These features with large dimensions (1024x1024) are then applied to Principal Component Analysis (PCA) to reduce data dimensionality and converted into 140x140 pixel size images. Finally, the extracted features are classified using the proximal support vector machines as classifier. The features with orientations of 0, π/4, 3π/4, and π/2 and also with all orientations of Gabor filters are combined with low frequency and high frequency filters and compared for Recognition rate are calculated. The Gabor filter with low frequency with all orientation gives the highest recognition rate.