Code structure is an important factor which affects performance of IRA codes. THe structure of IRA codes was optimized based on Gaussian approximation. The graph model theory of IRA codes was introduced, and the Tanner graph was presented. The impact of 2 degree to the structure of codes was analysed. Based on given noise variance, the Gaussian approximation of IRA codes was researched on AWGN channel, and the degree sequences was designed. IRA codes were designed based on Gaussian approximation and simulated by computer. Simulation results showed that the IRA codes have better performance than MacKay LDPC codes, and the convergence of message are well.