Abstract：To calculate the reliability index of the implicit nonlinear limit state equation, this paper proposes a structural reliability algorithm by combining the support vector machine (SVM) and particle swarm optimization (PSO). First, each iteration sample point is added to the sample points of this iteration as training samples for SVM, based on the advantage that SVM would not be limited by sample points. Secondly, the PSO method is introduced to calculate reliability index so as to solve the situation that the reliability index calculation of nonlinear limit state equation does not converge in the iterative process. Finally, importance sampling method is adopted to calculate failure probability according to the SVM regression model from the reliability index convergence. The numerical results illustrate that the failure probability calculated by this method has better precision and especially for the non-convergence of reliability index in the iterative process.