Cable-stayed grid structure is an original long-span spatial structure with modern prestressing techniques and cable-stayed bridge techniques. With good structural performance and original architectural shape, it can meet the requirements of modern architecture and is suit for modern sports, public and industrial building.The research work involves reliability calculation method, wind-induced analysis, sensitive analysis and reliability-based design method of the cable-stayed double-layer cylindrical latticed shell. A linear feasible direction algorithm is proposed to calculate reliability index. It overcomes the problem that the calculation of reliability index is probably not converged with the first order reliability method (FORM) when limit state equation is nonlinear. Based on the feasible direction strategy, the effect of objective function and limit state equation on calculation of reliability index is considered in the iteration of the algorithm. Numerical examples show that the algorithm is of higher precision and efficiency For the reliability index result with the second order reliability method (SORM) is sensitive to the design point solution, a method is proposed to improve reliability index based on the sampling point on the real limit state surface, which considers the geometrical characteristic of the limit state surface and approximate inaccuracy of the response surface method. Numerical examples show that the method is of high precision and not sensitive to the design point solution.Based on the previous research achievement, two improved response surface methods are proposed, called the state-space response surface method and the limit state response surface method. They provide a preliminary useful study for general case that how to apply the response surface method to analyze structural reliability problem when the performace function value is not used. The application of pattern recognition techniques and adaptive sampling techniques improves the reliability analysis efficiency. Numerical examples show that the two methods are of high precision with fewer finite element analysis times and suitable for the reliability analysis of the long-span spatial structures.Based on the auto-regressive (AR) method and the harmony superposition method, the programme on multi-nodal wind velocity time series simulation is compiled. Through the comparison of stimulant power spectrum and target power spectrum, the validation of the programme is proved. From the comparison it shows that the AR method simulates wind velocity fast but imprecisely, and the harmony superposition method simulates wind velocity accurately but time-consumingly. Then the wind-induced analysis is performed with the stimulant wind velocity time series.