A method for the characterization of flow-excited acoustic systems in the frequency domain was investigated. The approach, based on dimensional analysis and the hypothesis that the system may be described using a linear source-filter model, allows the source spectral distribution and the magnitude of the frequency response function to be obtained directly from sound-pressure spectra measured over a range of flow velocities. The objective as to determine whether the method could be used to predict the noise emissions of a small axial-flow cooling fan in a given installation from source and response functions evaluated using different calibration facilities. The spectral decomposition algorithms were first validated using a theoretical model. The influence of discretization errors and random errors in the input spectra was evaluated. The noise emissions of two different fans were then measured for two different computer enclosures and a calibrated plenum, following standardized procedures. The results demonstrated that the use of the spectral decomposition method as a pre-dictive tool was accurate in this case, provided the fan inflow was reasonably uniform. In addition, the method was found to be very robust to random errors in the input data. Discretization errors, however, had a significant impact on accuracy.