When building regression models, often we have enough knowledge to model some features of the mean response function parametrically, but only have vague knowledge about other features. Thus we want to leave these vague features unspecified and model them non-parametrically. A partial spline is an example where the mean function depends on both parameters and some non-parametric functions linearly. In this section we consider more general semi-parametric nonlinear regression (SNR) models where the mean function may depend on both parameters and non-parametric functions nonlinearly.