This paper presents a method for neural network ensemble. In the method, five subsystems of classifier used four kinds of neural networks, such as SOM,PNN,LVQ,RBF. Those neural networks compute parallel which have been trained solely The recognition result of subsystems, the expectation and variance between input pattern and goal pattern have been integrated by analytical hierarchy process. In experiments, the proposed methods have been successfully evaluated using thirteen different datasets, it is more effective than the relative majority voting scheme. The integration method consumes little computing resource and the result of calculation accords with the actual conditions, which indicates that AHP is an efficient ensemble method.