The simulation of the cyclone separation efficiency can rapidly become very expensive in time of calculation, which is not suitable for industrial application. The prediction of the complex turbulent gas flow inside a cyclone is necessary to estimate its pressure drop, its characteristics and finally its separation efficiency. Even though classic models such as RSM and LES can effectively predict such type of flow, the quality of the results and the computation cost can be discussed. In this work, Partially Averaged Navier–Stokes (PANS) method is proposed to simulate the fluid flow inside a gas cyclone. Compared to experimental and numerical results of a literature review, this model predicts reasonably the time-averaged tangential and axial velocity fields, and more accurately the pressure drop across the cyclone in shorter CPU times. The sensitivity of the PANS approach to the unresolved to total kinetic energy ratio f k was also evaluated with two different definitions. The first one was the standard definition developed by Girimaji et al. [1] and the second one was based on a statistical equivalence between the DES and PITM. The proposed model offers a better cost to quality ratio compared to the classic RANS approach.