This paper presents appearance based methods for face recognition using linear and nonlinear techniques. The linear algorithms used are Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). The two nonlinear methods used are the Kernel Principal Components Analysis (KPCA) and Kernel Fisher Analysis (KFA). The linear dimensional reduction projection methods encode pattern information based on second order dependencies. The nonlinear methods are used to handle relationships among three or more pixels. In the final stage, Mahalinobis Cosine (MAHCOS) metric is used to define the similarity measure between two images after they have passed through the corresponding dimensional reduction techniques. The experiment showed that LDA and KFA have the highest performance of 93.33 % from the CMC and ROC results when used with Gabor wavelets. The overall result using 400 images of AT&T database showed that the performance of the linear and nonlinear algorithms can be affected by the number of classes of the images, preprocessing of images, and the number of face images of the test sets used for recognition.