An automated procedure for performing large-scale sensitivity studies based on the use of computer calculus is presented. The procedure is embodied in a FORTRAN precompiler called GRESS, which automatically processes computer models adding derivative-taking capabilities to the normal calculated results. The theory and applicability of the GRESS code are described and tested against a major geohydrological modeling problem. The SWENT nuclear waste repository modeling code is used as the basis for these studies. Results for a test problem involving groundwater flow in the vicinity of the Richton Salt Dome are discussed in detail. Sensitivity results are compared with analytical, perturbation, and alternate sensitivity approaches to the problem. Five-place accuracy in these sensitivity results is verified for all cases in which the effects of nonlinearities are made sufficiently small. Conclusions are drawn as to the applicability of GRESS in the problem studied and for more general large-scale modeling sensitivity studies.