A face recognition method is proposed based on multi features description and local fusion decision. Firstly, we use Independent Component Analysis (ICA) to construct the global complementary subspace to roughly classify the test samples. Then the texture descriptor algorithms under three different definitions are used to construct local complementary subspace to obtain the posterior probability of sample which is difficult to classify by rough classification. Finally, we get the precise classification result of test sample on the local complementary subspace through setting grade scores based on the value of the posterior probability. The experimental results on ORL, Yale and FERET face database show that the proposed method better describes characteristics of the image and has lower time complexity and higher recognition rate. Compared with other methods, it also proves its effectiveness on the face recognition.