The layering in reservoir simulation grids is often based on the geology, e.g., structure tops. In this paper we investigate the alternative of using horizontal layers, where the link to the geology model is by the representation of the petrophysics alone. The obvious drawback is the failure to honor the structure in the grid geometry. On the other hand, a horizontal grid will honor the initial fluid contacts perfectly, and horizontal wells can also be accurately represented. Both these issues are vital in thin oil-zone problems, where horizontal grids may hence be a viable alternative. To investigate this question, a number of equivalent simulation models were built for a segment of the Troll Field, both geology-based and horizontal, and various combinations of these. In the paper, it is demonstrated that the horizontal grid was able to capture the essentials of fluid flow with the same degree of accuracy as the geology-based grid, and near-well flow was considerably more accurate. For grids of comparable resolution, more reliable results were obtained by a horizontal grid than a geo-grid. A geo-grid with local grid refinement and a horizontal grid produced almost identical results, but the ratio of computing times was almost 20 in favor of the horizontal grid. In the one-phase regions of the reservoir, relatively coarse cells can be used without significant loss of accuracy.