Two different architectures of Independent Component Analysis (ICA) to palmprint recognition which were architectureⅠand architectureⅡwere discussed.Region of Interest (ROI) in the palmprint images were extracted automatically by preprocessing before feature extraction and match so as to increase the recognition accuracy and reliability.In order to reduce computational complexity,Principal Component Analysis (PCA) was used to eliminate second-order dependencies in the palmprint images.The remaining higher-order dependencies were separated by ICA.The Square Project Error (SPE) of ICA model was smaller than that of PCA,and the reconstruction of the original palmprint images was superior to that gotten by PCA in PolyU Palmprint Database.To compare the recognition performance of two ICA architectures with PCA,we applied them to extract the palmprint feature subspace inside ROI.Then the images to be recognized were projected on small dimension subspace.Finally,we used a classifier to palmprint match based on cosine distance.Experimental results show that two ICA architectures perform better than PCA and ICA architectureⅡis the best in performance.