The acoustic emission of radial fans is an important issue in product development. In addition to experiments in acoustic test rigs numerical methods represent alternatives to predict sound emission and to gain insight into sound generation mechanisms. In this project a backward curved radial fan is analysed in two configurations: fan alone and fan installed in spiral casing. Experimental investigations were done at the anechoic fan test rig at TU Kaiserslautern. The experimental data was used to validate CAA Simulations based on finite volume LES. Two methods were used for the calculation of sound at the microphone positions: the direct evaluation of the compressible pressure and far field pressure evaluated with the FW-H (Ffowcs Williams and Hawkings) equation. The FW-H integral equation assumes free-field radiation from the sound emitting surfaces. The method is applied for both configurations, although assumptions are not valid the fan in spiral casing. In contrast the direct method requires sufficient mesh resolution in the propagation zone in order to resolve the sound waves. Mesh size studies were done to investigate the required number of cells per wave length. A 3rd order scheme for discretization of the convective terms was applied, which resulted in very good agreement of measured acoustic spectra with 10 to 15 cells per wave length. In case of the fan alone a periodic simulation model was applied in comparison to a full rotor simulation. Good agreement of FW-H and direct acoustics were observed. The limits of the periodic approach were investigated. The approach is not valid at operating points with large unsteady separations. In addition the periodic simulations result in increased levels due to the coherent sound field. On the axis of rotation a simple correction of the sound pressure level is applicable to obtain the physical uncorrelated sound emission. For pressure probes off-axis increased levels compared to full model simulations were observed. This is especially observed for low frequencies.