The dependence of van der Pauw resistivity measurements on local macroscopic inhomogeneities is shown to be nonlinear. A resistor grid network models a square laminar specimen, enabling the investigation of both positive and negative local perturbations in resistivity. The effect of inhomogeneity is measured both experimentally, for an 11x11 grid, and computationally, for both 11x11 and 101x101 grids. The maximum "shortlike" perturbation produces 3.1 +/- 0.2 times the effect predicted by the linear approximation, regardless of its position within the specimen, while all "openlike" perturbations produce a smaller effect than predicted. An empirical nonlinear correction for f(x,y) is presented which provides excellent fit over the entire range of both positive and negative perturbations for the entire specimen. (c) 2006 American Institute of Physics.