Recently, multiple classifier systems (MCS) have been used for practical applications to improve classification accuracy. Self-generating neural networks (SGNN) are one of the suitable base-classifiers for MCS because of their simple setting and fast learning. However, the computation cost of the MCS increases in proportion to the number of SGNN. In this paper, we propose a novel optimization method for the structure of the SGNN in the MCS. We compare the optimized MCS with two sampling methods. Experiments have been conducted to compare the optimized MCS with an unoptimized MCS, the MCS based on C4.5, and k-nearest neighbor. The results show that the optimized MCS can improve its classification accuracy as well as reducing the computation cost.