This paper presents a new adaptive sampling approach based on a novel integrated performance measure approach, referred to as "iPMA," for system reliability assessment with multiple dependent failure events. The developed approach employs Gaussian process (GP) regression to construct surrogate models for each component failure event, thereby enables system reliability estimations directly using Monte Carlo simulation (MCS) based on surrogate models. To adaptively improve the accuracy of the surrogate models for approximating system reliability, an iPM, which envelopes all component level failure events, is developed to identify the most useful sample points iteratively. The developed iPM possesses three important properties. First, it represents exact system level joint failure events. Second, the iPM is mathematically a smooth function " almost everywhere." Third, weights used to reflect the importance of multiple component failure modes can be adaptively learned in the iPM. With the weights updating process, priorities can be adaptively placed on critical failure events during the updating process of surrogate models. Based on the developed iPM with these three properties, the maximum confidence enhancement (MCE) based sequential sampling rule can be adopted to identify the most useful sample points and improve the accuracy of surrogate models iteratively for system reliability approximation. Two case studies are used to demonstrate the effectiveness of system reliability assessment using the developed iPMA methodology.