The development of coarsened reservoir simulation models from high resolution geologic models is a critical step in a simulation study. The ‘optimal' coarsening becomes particularly challenging in a fluvial channel environment where the channel sinuosity and orientation can result in pay/non-pay juxtaposition in many regions of the geologic model. Under such conditions, a uniform coarsening will result in mixing of pay and non-pay and will most likely result in geologically unrealistic simulation models and erroneous performance predictions. In particular, the upgridding algorithm must keep pay and non-pay distinct through a nonuniform coarsening of the geologic models. In this paper we present a coarsening algorithm to determine an ‘optimal’ reservoir simulation grid by grouping fine scale geologic model cells into effective simulation cells. Our algorithm groups the layers in such a way that the heterogeneity measure of an appropriately defined ‘static’ property is minimized within the layers and maximized between the layers. The optimal number of layers is then selected based on an analysis resulting in the minimum loss of heterogeneity because of upgridding. We demonstrate the application of the optimal gridding by history matching waterflooding in a structurally complex and faulted offshore turbiditic oil reservoir. The field is located in a prolific hydrocarbon basin offshore South America. Permeability and fault transmissibility are the main uncertainties. More than 10 years of production data from up to 8 producing wells are available for history matching. We demonstrate that any coarsening beyond the degree indicated by our analysis overly homogenizes the properties on the simulation grid and alters the reservoir response.