A neural network for the implementation of mean-gain-shape vector quantization is proposed. Mean-gain-shape vector quantization is a product vector quantization consisting of three codebooks. A counterpropagation network (CPN) is used to perform the vector quantizations. The CPN is a combination of two well-known algorithms: the self-organization map of Kohonen and the Grossberg outstar. The proposed approach is more efficient than the conventional LBG algorithm in terms of computational complexity. Moreover, the issue of optimal bit allocations is studied through extensive experimentation and interesting results are obtained.