After the number of 'important' factors has been reduced to two or three, it is possible to investigate in more detail how these factors affect the response. When the factors are numerical, it is possible to model the mean response with a nonlinear function of the factors. The experimental design must involve three or more levels of each factor in order to fit a model that has quadratic terms in all factors.

If the factors are proportions of ingredients that must add to one, a similar quadratic model is used, but the experiment must be designed in a different way.