Computational cell size is a crucial factor for accuracy in Computational Fluid Dynamics (CFD) - Discrete Element Method (DEM) simulations of particle-fluid interactions. In the present study, we investigate how simulation results change with computational cell size and mixture composition, for calculation of drag force over a fixed bed containing a binary-sized particle mixture. To investigate the complex solution convergence behavior, the simulation results are examined for several definitions of dimensionless computational cell size. Several regimes of consistent behavior, across three investigated mixtures, are identified and a consistently optimal cell size range is identified. We find that both the difference between simulated solution results and published experimental results, and the standard deviation of the void fraction profile, show consistent trends when plotted against the dimensionless computational cell size based on the Sauter-mean particle diameter. Grid-refinement studies are performed across all grid solutions, and the Grid Convergence Index (GCI) is analyzed as a predictor for the grid solution error. Correlations between simulation error and GCI are not strong, likely because of incongruence of the solution trends with typical asymptotic convergence. Alternatively, a correlation between change in solution value on successively refined grids and finer-grid solution error is shown to be adequate for the current results. (C) 2017 Elsevier B.V. All rights reserved.