Establishing precise mechanical models of magnetorheological dampers is an important prerequisite not only for the response analysis and design of structures with the dampers, but also for obtaining good vibration control effect. In this paper, the dynamic performance test of a magnetorheological damper with the nominated maximum damping force of 10 kN is first carried out. Based on the test results, parametric and non-parametric dynamic models of the damper are developed and the effectiveness of the proposed models is verified. Finally, the results of the two different modeling approaches are compared. Results show that the formulated hyperbolic tangent hysteresis model (parametric model) could well describe the dynamic behavior of the magnetorheological damper. Besides, both the forward and reverse back propagation (BP) neural network models (non-parametric models) have good performance in fitting training data, generalization ability and noise immunity. Furthermore, the BP neural network models fit with a higher accuracy than the hyperbolic tangent hysteresis models, while the later enjoy a simpler expression of the damping force which is relatively easier to implement in software.