The lubricated clutch in an automatic transmission plays an important role in the performance and comfort of passenger vehicles. Therefore, an accurate and easy-to-implement dynamic clutch model is necessary for powertrain system design and performance studies. A neural network approach recently developed by Parvataneni et al.  for clutch modeling has illustrated some very promising results. However, this model has complex architecture that may cause slow training and testing. Also, due to the lack of time information, the network cannot adapt to time step variations. Therefore, it cannot be easily integrated with powertrain system models, which in general require variable time steps for a superior numerical integration performance. In this paper, a new first-principle-based hybrid network clutch model is derived for dynamic engagement analysis with variable time steps. With improvement over the previous work by Parvataneni et al. , the time pattern information is added to the inputs and a simpler architecture is developed through more explicit utilization of the physical laws. A second order training algorithm with dynamic derivatives is also used to improve the training efficiency and accuracy. With these new features, this model can significantly outperform the previous approach in terms of accuracy and efficiency. The network is trained and tested using experimental data as well as analytical results. It is shown that this new model can compensate for time step variations and can predict the clutch torque accurately for a wide range of operating conditions.