The hydrocyclone has been widely used in the mineral industry for over hundred years, yet the standard geometry of hydrocyclones has remained almost unchanged. The exploration of new designs is time-consuming and costly to do by experimentation. In this paper, the computational fluid dynamics tool is used to explore alternative geometries in a way to manipulate the hydrodynamics to achieve the desired classification. Fluent (TM) 6.0 was used to solve the governing equations. The large-eddy simulation model was used for the turbulence closure and the Lagrangian particle-tracking method was used to predict the particle classification. Six new geometries are explored and compared with the standard design. The mass balance and the classification curve are the variables used to evaluate the performance of each of the novel designs. The results show that the modification of the geometry in designs #1 and #2 did not improve the classification performance, and they are not suitable for experimental validation. Designs #4 and #5 showed a similar performance achieved with the standard designs, whereas designs #3 and #6 were the most promising configurations to create a sharper particle classification. Designs #3 and #6 will be excellent candidates for further experimental validation. The principal contribution of this paper is that the computational fluid dynamics is the right tool to study and explore novel designs of hydrocyclones. (c) 2006 Elsevier B.V. All rights reserved.