Unimodal biometrics is easily affected by conditions and environmental variations. Since face recognition and fingerprint recognition have been widely used in biometric systems, a multimodal biometrics method was presented based on the fusion of face and fingerprint features. The image processing methods were applied to the face and fingerprint images, and both LBP and Gabor features were extracted for these two biological features. Then, a generalized canonical correlation analysis was introduced into the multi-feature fusion of faces and multi-feature fusion of fingerprints respectively. A block diagonal matrix was taken to combine the above fused face features and fingerprint features. Finally, a classification was executed by solving the robust probabilistic collaborative representation. The experimental results on two multimodal databases demonstrate that the multimodal biometrics of faces and fingerprints has higher accuracy and better stability than the unimodal biometrics. The proposed fusion method based on the generalized canonical correlation analysis is superior to traditional methods.