Abstract: Using the ear and face possessing of special physiological correlation under the same pose condition as the research object, a muhimodal recognition method based on kernel canonical correlation analysis (KCCA) was proposed to solve the problem of information loss resulted from sharp pose change. In the method, the normalization and centering methods were used to preproeess ear and face datasets and the nearest neighbor method was used to classify. Experimental results show that KCCA can availably overcome the effect of sharp pose change. Compared with the single biometric, the recognition rate improves remarkably.