The dynamics of hydrocyclones is complex, because it is a multiphase flow problem that involves interaction between a discrete phase and multiple continuum phases. The performance of hydrocyclones is evaluated by using Computational Fluid Dynamics (CFD), and it is characterized by the pressure drop, split water ratio, and particle collection efficiency. In this paper, a computational model to improve and evaluate hydrocyclone performance is proposed. Four known computational turbulence models (renormalization group (RNG) k- $\epsilon$ , Reynolds stress model (RSM), and large-eddy simulation (LES)) are implemented, and the accuracy of each for predicting the hydrocyclone behavior is assessed. Four hydrocyclone configurations were analyzed using the RSM model. By analyzing the streamlines resulting from those simulations, it was found that the formation of some vortices and saddle points affect the separation efficiency. Furthermore, the effects of inlet width, cone length, and vortex finder diameter were found to be significant. The cut-size diameter was decreased by 33% compared to the Hsieh experimental hydrocyclone. An increase in the pressure drop leads to high values of cut-size and classification sharpness. If the pressure drop increases to twice its original value, the cut-size and the sharpness of classification are reduced to less than 63% and 55% of their initial values, respectively.