In structural reliability In reliability analysis of structure, the limit state function is often implicit and nonlinear influenced by many factors, usually using the response surface simulation. And the structure of the small failure probability is difficult to get a large number of reliability data to verify the correctness of the response model. The form of polynomial limits the accuracy of reliability calculation.Therefore the optimization kriging model method is put forward to improve the precision, optimize deviation and reduce the sample. A function of variables and the response is obtained by the approximate model, which replace the limit state function in reliability analysis. But the kriging model requires a lot of sample verification, global optimization(GO) is used to determine the sample points near the limit state and relatively high uncertainty as the new sample. And so a less sample is used to build kriging surrogate model meeting precision, and then the structural reliability is computed gradually with H-L(Hasofer-Lind) method. The numerical simulation results show this method is the higher accuracy. Comparing with the classical kriging, MCS and FOSM, the method is less computation error and less sample. It can meet the engineering demand and greatly reduce the amount of calculation..