Abstract: There are usually several failure models in a complex structural system. The conventional model for system analysis is built under the condition that all failure models are independent with each other. However, in engineering practice, failure models in a system are mostly dependent due to the fact that the elements of a structural system are closely interrelated. System reliability analysis and evaluation simply conducted under the condition that all failure models are independent with each other often result in excessive errors or even wrong conclusion. This paper proposes a method for reliability analysis of engineering structural systems with dependent failure models. The dimensional reduction method and Gauss-Hermite quadrature are applied to compute the statistic moments of the limit state functions of structural systems with random parameters. The first four cumulants of the limit state functions are used to approximate cumulant generating functions. The probability density function and cumulative distribution function of the limit state functions of the structural systems are fitted through saddlepoint approximation. And then the reliability (or failure probability) of the structural system is obtained. Finally, the practicality and efficiency of the proposed method are demonstrated by a numerical example.