The Reflux (TM) Classifier (RC) is the state-of-the-art liquid-solid fluidized bed gravity separator and its industrial acceptance and commercialization have been increasing in the coal preparation. Despite the significant advancement of this technology, optimization studies, which could be of paramount interest to the plant operator, were not found in the previously published literature. The objective of this research work is to develop a statistical model to predict the ash content and yield of the concentrate. Box-Behnken design of experiment was employed to analyze the gravity separation performance of the intermediate size coal fraction (0.25 x 2 mm). Three important operating parameters (bed setpoint, feed rate and fluidization water rate) were considered to study the separation efficiency of RC. To achieve minimum ash content with the maximum yield of concentrate, multi response optimization was carried out using desirability approach to optimize the concentrate ash (%) and yield (%) simultaneously. The global optimum yield value of 50% concentrate was achieved with an ash content of 11.5%, at the bed density setpoint of 1100 kg/m(3), feed rate of 5.9 l/min and water fluidized rate of 4.1 l/min. The influence of operating variables of the RC on ash content and yield of concentrate are presented and discussed in 3D surface plots. The developed statistical models for the ash content and yield of concentrate when tested were found to predict the yield with an average error of 5%. Partition curves showed that the separation density (rho(50)) varied from 134 to 1.52 and the Ecart probable error (Ep) varied from 0.075 to 0.1 for the Reflux (TM) Classifier. (C) 2019 Elsevier B.V. All rights reserved.