Aiming at the characteristics of strong subjectivity and weak adaptability of traditional test equipment maintenance support capability evaluation, a maintenance capability evaluation method based on error back propagation （BP） feedforward neural network is proposed based on test equipment maintenance records as sample data. In order to eliminate the attribute redundancy of the maintenance records， the rough set principal component reduction algorithm is selected to reduce the sample attributes. In order to avoid the influence of the local convergence and the slow convergence speed of the BP neural network due to the local extremes,the initial weights and thresholds of neural network are optimized by using the global search ability of genetic algorithm (GA). In order to increase the learning function of the network and improve the network training rate and training accuracy， the wavelet function is selected as the transfer function of the hidden layer， and the Sigmoid function is the activation function of the output layer. It is shown that the performance of the BP neural network is improved through redundant attribute reduction， and the weight/threshold value of the BP neural network and the wavelet function are optimized by GA as the transfer function of the hidden layer， which avoids local convergence and improves the training rate of the network. The accuracy of the evaluation reduces the error.