As for the existing neural network algorithm，a new kind of optimized BP neural network algorithm is put forward and applied to network quality evaluation. In the optimization scheme, the stages of rise and fall are optimized by adopting different strategies.The theoretical analysis shows that the use of the optimized algorithm can overcome the previous shortcomings of slow weight convergence and shaking problem in error convergence .The experiments show that the use of the optimized algorithm will decrease the learning error of neural network and the quality classification error obviously, and simultaneously improve the accuracy of the evaluation significantly. The error of the optimized algorithm is more stable than that of the traditional algorithm during the process of convergence. As the result, the learning error falls by 9.64% and the decline of quality classification error is 23.1%. Compared with the traditional and step size variable algorithm, the use of the optimized algorithm raises the checking accuracy rate separately by 19.65%and 9.88%. It's obvious that the optimized algorithm is effective in raising the prediction accuracy of the BP neural network by a large margin.