The paper proposes a novel face recognition technique based on discrete Contourlet Transform (CT). CT represents smooth contour information in different directions and so relevant to recognize individuals more accurately. Each face image is decomposed up to fourth level in transformed domain using Contourlet Transform and coefficients are analyzed to obtain statistical features. For classification feature vectors are obtained considering training images of each database. To reduce dimension of the feature vectors, directional subbands are selected by analyzing the entropy of the feature vectors. For the recognition of face images Support Vector Machine (SVM) is used. The experimental results give promising performance of the proposed face recognition method on JAFFE and ORL database compare to other transformed domain methods.