A reliability calculation method is proposed to calculate the reliability of structure with complex implicit function like steel truss structures. It firstly adopts the neural network to approach the implicit function in service ability limit state and NGA (New Genetic Algorithm) is employed to obtain the optimal solution of reliability index of steel truss structures and its design point on the basis of the geometric implication of reliability index. Finally，JC method and Monte Carlo Critical Sampling Method are introduced，respectively，in two examples to verify the accuracy and validity of NGA. The results manifest that the relative error is only 0.23 percent when NGA and Monte Carlo Method are used，respectively，to calculate the reliability index of steel truss. In addition，the introduction of adaptive random variable can greatly improve the gene of initial population for small probability failure structures. All above prove that NGA is of significance in practical projects for calculating the reliability index of structure with complex implicit function due to its advantages of fast computation speed and high precision.