To study dynamic evolution law of mechanical reliability caused by wear, gear transmission system is taken as a research object. Considering the effect of clearance caused by wear on gear teeth load in double meshing area, the formula of dynamic distribution load which is undertaken by two adjacent teeth is deduced. And the distributed pressure and meshing speed, which should be taken into account while calculating gear wear, are obtained based on the Winkler surface model and principle of tooth mesh. Based on the Archard’s wear model, numerical simulation model for wear in spur gear is deduced, and the wear depth of each meshing points on teeth outline with different wear cycles are obtained. The calculation wear model is replaced with a surrogate model with Neural Network and Kriging method to overcome time-consuming defect. Random process model is integrated with the surrogate model, and dynamic reliability for nonlinear stochastic structure with unknown distribution characteristic is obtained with Neural Network-based Edgeworth series technique and four moment methods, which is compared with Kriging-based Monte Carlo simulation method. The computational efficiency and accuracy are also demonstrated.