The aim of this study was to develop an efficient probability-based seismic fragility assessment method for steel moment-resisting frames (MRFs). In this method, Bayes estimators are used to improve the accuracy of the results based on new analyses or experimental observations. The efficiency of Bayes estimators in the updating process of fragility curves is demonstrated based on the results of incremental dynamic analysis of three-, five- and seven-storey steel MRFs under a set of 40 spectrum-compatible earthquake ground motions. The results indicate that the constant-hazard-level (engineering demand parameter based) fragility analysis approach leads to less than 6% error in the prediction of performance levels and therefore can be considered a competent alternative to the more computationally expensive constant-performance (intensity measure based) approach. While the earthquake duration can considerably affect the results of fragility analysis, the prediction errors can be reduced to an acceptable level through selection of four different time intervals with at least three records in each group. It is also concluded that Bayes estimators can generally provide a very good level of accuracy in updating fragility curves by reducing the maximum errors to less than 2.5% compared with the exact values.