Fine grid geocellular reservoir characterizations often include detailed description of geological and geometric complexity. Since this volume of details cannot be handled by commercial reservoir simulators, some degree of coarsening is necessary. The fundamental agenda in coarse grid generation includes development of finer grids in regions with higher flow. Available methods in literature suffer from draw back of inability to consider both the well and geological features which mostly affect the flow response. This study introduces a new method of grid coarsening (flow based) procedure. The procedure encompasses tracing streamline from boundary to the well, monitoring velocity trend along streamline to identify high flow region and selection of appropriate points on streamlines as grid nodes. Differentiating the analytical equation of the streamline path results in developing velocity vectors and adding them up will yield a new term called cumulative velocity. Using this term, the grid points are easily identified which after implementation of Delaunay triangulation and Laplacian smoothing algorithms the main CVFE (Control Volume Finite Element) grid is produced. In addition, a robust up-scaling technique was used to calculate the tensor of permeability. Flux-continuous finite volume scheme was used to solve the associated flow equation in the coarse block. The generated grid pattern is finer in high flow regions and can successfully adapt itself based on type of geological features presented. Pressure was the main criteria for comparison the results of this study with those of Cartesian coarse grid. The results indicated that the response of the model with coarse grids (flow based) is more consistent with a fine model. The major advantage of the introduced method is its capability in handling the geological features within the grid. It is worthy to mention that the positioning of the well and adjustment of the required refinements around it are done automatically. Introduction Reservoir simulation is an important tool for the management of oil fields. Geological characterization and equations governing flow through porous media is used to simulate reservoir flow numerically. But where to consider geological variations and how much detail to consider when dealing with them, are questions that challenge engineers. During the process of griding and discritization physical properties are assigned to each grid block. Concentration of grid blocks represents the degree of detail we have considered and its location in the domain shows where geological variations have more importance. This indicates that grid generation method plays an important role when regarding number of grid produced. Geological models contain as much as grid blocks that cannot be handled by commercial reservoir simulators software due to computational cost. These models are called fine grid models or geostatistical models which contain fine scale geological variation. In order to reduce the number of blocks we should use coarser blocks. But petrophysical properties like permeability are defined in fine grid model. There should be a process to assign fine grid properties to the coarse grid. This process is called upscaling. In the process of griding and upscaling the tendency is to preserve critical reservoir features as much as possible. In this paper, we tried to take into account this fact by concentrating grid nodes in important regions to capture reservoir features. Different attempts have been performed so far in this area. The first one was the idea of local grid refinement (Ciment and sweet 1973; Pedrosa and Aziz 1985). Local grid refinement involves using fine grid inside coarse base grid. This is done in selected regions like near well regions or highly heterogeneous regions.