For a class of nonlinear systems, the residual of a well fitted model has low intrinsic dimensionality. For these systems, a particular low-dimensional linear projection of the regressor will facilitate both visualization of the nonlinearities and subsequent nonlinear modeling. The least squares fit of polynomial and piecewise affine functions are used as criterion by which numerical programs search for the linear projection that gives the best low-dimensional description of the residual. For a simulated water tank and for real life data sampled from an electronic circuit, the regressor can be projected down to 2 dimensions and still yield a model simulation fit of about 99%. The electronic circuit data can be described by a model structure with far less parameters than conventional, nonlinear black-box models.