讨论了线性分组码互补码的特性及分解，提出了基于神经网络的互补码分解译码方案，此方案利用神经网络的吸引子和吸引域进行纠错译码，并把互补码分解为子码及其若干陪集，更进一步减小神经网络译码的复杂性及规模，从而实现高效实时硬判决译码．给出了实现原理及步骤，并对其译码性能进行了分析比较． This paper discusses the characteristics and decomposition of linear block codes, and proposes a complementary code decomposition decoding scheme based on neural network. This scheme uses the attractor and the attracting domain of neural network to carry out error correction decoding and decomposes the complementary codes into sub codes And some of its coset, to further reduce the complexity and size of the neural network decoding, in order to achieve efficient real-time hard-decision decoding. The implementation principle and steps are given, and its decoding performance is analyzed and compared.