Rapid seismic fragility analysis of regular bridges is necessary for the increasingly utilized regional seismic risk assessment, which is usually difficult using numerical methods owing to the computational requirements. Therefore, this study presents an artificial neural network (ANN)-based methodology for regular multi-span bridges, which considers the influencing characteristics of bridges and their uncertainties. For the ANN model development, finite element (FE) models and fragility analyses of regular bridges are investigated in detail, based on which the influencing characteristics and the fragility parameters of critical components are identified as the ANN inputs and outputs, respectively. In addition, a bridge design procedure that integrates the seismic codes and engineering experience is implemented to automatically generate detailed FE models of bridges according to the input characteristics. Next, a well-distributed bridge database of 516 bridges is designed, and incremental dynamic analysis (IDA) is conducted, yielding the fragility results. Finally, the ANN-based fragility model is trained and generated, and its results are compared with those based on IDA. The good agreement (root mean absolute error of 0.173 and coefficient of determination of 0.997) indicates that the proposed method is an effective alternative for seismic assessment of bridges with significantly reduced computation time.