Finite-difference reservoir simulators are potentially a powerful and versatile tool for analyzing field tracer data to infer an accurate reservoir description. However, to realize this potential the simulator must be physically and numerically accurate and computationally affordable. We have previously reported on the simulation of a large tracer field test in a very heterogeneous sandstone oil reservoir under waterflood. The mesh used in this study was relatively coarse to keep the computational times down for the very large number of simulations that were required to carefully analyze these tracer data. In this paper, we present the results of a mesh refinement of the final history match of these data that shows the original grid was satisfactory and thus validates these results. These very large simulations were made feasible by the use of microtasking on a four-processor CRAY 2. A higher-order finite-difference method for the convective derivative was used to reduce both truncation and grid-orientation errors in these simulations. Thus, we have established that with this method accurate simulations of field tracer data are affordable.