This paper considers the problem of model selection and model averaging for regression models which can be nonlinear in their parameters and variables.We propose a new information criterion called nonlinear model information criterion (NIC),which is proved to be an asymptotically unbiased estimator of the risk function under nonlinear settings.We also develop a nonlinear model averaging method (NMA) and extend NIC to NICMA criterion,which is the corresponding weight choosing criterion for NMA.By taking account of the complexity of model forms into the penalty term,NIC and NICMA achieve significant gain of performance.The optimality of NMA,convergence of the selected weight and other theoretical prop-erties are proved.Simulation results show that NIC and NMA lead to relatively lower risks compared with alternative model selection and model averaging methods under most situations.