In order to solve the robustness issues for single training sample face recognition under occlusion conditions, this paper presents a recognition method based on adaptive weighting and fuzzy fusion. In our method, the information entropy expansion mode is introduced in sub-mode method and via adaptively assigning the weights corresponding to each sub-model can reduce the impact of occluded region. In addition, through summing the similar blocks in face images can make up for the defect of the sub-model which ignores the integrity of face. Finally, the method based on fuzzy comprehensive evaluation is utilized for decision-level fusion to the outputs by these two ways of classification. Experimental results on AR face database show that this method has state-of-the-art classification accuracy and also robustness to occlusion.