Various computational fluid dynamics (CFD) models employed for cavitating flows are substantially based on semi-empirical assumptions about cavitation forms and liquid flows around cavitating bodies. Therefore, the model applicability must be validated with experimental data. The stages of validation of the models are analyzed here with data on cavitating hydrofoils and axisymmetric bodies in water. Results of Reynolds-averaged Navier–Stokes (RANS), large-eddy simulation (LES), detached-eddy simulation (DES), and viscous-inviscid interaction methods are compared. The necessity of simultaneous validation of several flow parameters (as cavitation inception number and location of the appearing cavity) is pointed out. Typical uncertainties in water tunnel model test data (water quality, simplified account of wall effect) and possibilities to take them into account are also discussed. The provided comparison with experimental data manifests the impossibility to describe initial stages of cavitating flows using any single model and importance of employment of a combination of models for both the cavitation zones and the flow outside of cavities.