A new CFD model for Liquid-solid Fluidized Bed Separator (LSFBS), namely the Eulerian-Eulerian-Lagrangian/RNG k-epsilon approach, was put forward and validated. According to the simulated separation results of two cases, the root-mean-square error (RMSE) between predicted partition numbers of all density fractions and corresponding experimental values were 2.47 and 2.83 respectively, demonstrating the CFD model was able to give accurate simulation results. The Rosin-Rammler model was used to describe the size distribution characteristics (SDC) of the feed, with the parameter D-x and n describing the fineness and particles size variation of the feed respectively. Single-factor tests and central composite design were then developed and simulated using the CFD model to investigate the influence of the two aspects of SDC on the separation performance of LSFBS based on density. The simulated separation results indicate that the finer the feed is, the greater the separation density (delta(50)) and E-p value are; the smaller the particles size variation is, the smaller the delta(50) and E-p value are. According to the response surface analysis, the influence of feed fineness on delta(50) is larger than that of particles size variation while the influence of feed fineness on E-p is smaller than that of particles size variation; the interactive effect between these two factors has a noteworthy influence on E-p but insignificant influence on delta(50). (C) 2016 Elsevier B.V. All rights reserved.