Optimisation of flow in centrifugal pumps is often performed using Computational Fluid Dynamics (CFD). With the aid of CFD, pump geometry can be optimised without using expensive experiments. However, numerical simulations are an approximation of a real flow and an error in results is always present. Critical part of fluid flow assumption in pumps is the rotor-stator interface treatment which is used to couple rotor and stator zone into a single domain on a matrix level. Selection of the rotor-stator interface treatment used in flow simulation effects accuracy and time required to perform numerical simulation. In this thesis, four different types of simulations are performed and compared: - multiple frames of reference (MRF) with general grid interface (GGI) between rotor and stator zone, -multiple frames of reference with mixing plane interface between rotor and stator zone, - transient simulation with dynamic mesh and time averaged results and - harmonic balance method. The multiple frames of reference model is a steady state approximation of a transient flow, which enables a steady state simulation of turbomachinery by ignoring transient phenomena. Rotation of the rotor in the MRF is taken into account with additional terms in mathematical model, such as centrifugal and Coriolis term, without a dynamic mesh. The harmonic balance method is a quasi steady-state model which enables calculation of results in specific time instances of a transient flow using a steady-state formulation. This is only possible with the assumption that the flow is periodic and that each primitive variable can be accurately represented by a Fourier series in time. Due to complex geometries of turbomachinery numerical spatial domains are often made from more parts merged together. This often leads to non-conformal cell zones of rotor and stator. In this thesis, interfaces used for joining multiple non-conformal domains are GGI and mixing plane interface. Introduced numerical models are applied on two centrifugal pump geometries. Q - H and Q - h performance curves are approximated by a parabola defined with three operating points. CPU time analysis is also covered. Additional methods, which are not specified in the assignment of the thesis, are done to compare additional numerical models. Comparison between consistent and non-consistent simulation shows that the consistent transient solver gives the solution with less oscillations. Comparison between coupled and segregated solver shows that the coupled solver, even though it requires more memory, improves the convergence of the results. Therefore with the development of new methods for simulations will in future converge faster, yielding lower CPU time.