In response to the problem that the current specifications are difficult to reflect the actual bridge safety due to the frequent occurrence of overweight vehicle loads in China， this paper constructed a framework for generating traffic load response considering the impact effect based on measured data from a weight-in-motion （WIM） system. The response of random traffic flow was extrapolated using the compound extreme value theory， and a model for traffic load effect probability that considers the influence of the evaluation benchmark period was established. Moreover， a fatigue reliability assessment method for cable-stayed bridges based on measured traffic flow was proposed. Taking a cable-stayed bridge with the measured data of three mouths from a WIM system as a background， the reliability of the cables was evaluated. The results show that， in the case of using the load probability model extrapolated based on the measured WIM data， the fatigue reliability index of the cables ranges from 3.93 to 6.12， and the reliability index of the ultimate limit state strength for carrying capacity ranges from 7.74 to 11.04. The fatigue reliability index ranges from 4.35 to 6.18 when using the standard load probability model， which indicates that the fatigue reliability is significantly lower than the strength reliability， indicating that the safety of the cables in the assessment benchmark period is mainly controlled by fatigue. Compared with load benchmark periods extrapolated based on measured WIM data， the standard loads will significantly overestimate the fatigue reliability of the cables.