We describe in this paper a method allowing to order submodels in linear regression. A real function is attached to each submodel, allowing to graphically compare and order them. Our procedure defines an objective function depending on two factors (lack of fit and multicolinearity) with the...