Cavitation tests using models are essential to determine the full-scale machine cavitation behaviour. As erosion prediction is still very empirical, in practice a "safe" machine shows no visible cavitation on a model at sigma plant. However, this way of defining the setting level can lead to errors due, among other things, to the nuclei transit time in the runner low pressure zone. It is shown in this paper that the absence of visible cavitation on a model does not correspond to absence of cavitation on the full-scale machine. Pressure distribution plays an essential role. Moreover, it is shown that not respecting Froude similarity can also lead to a false determination of the full-scale machine setting level, as this similarity is the only one which guarantees similar pressure distributions between model and prototype. As a correlation between cavitation and water quality is essential to give a physical explanation to the results, measurements of the water nuclei distribution must be made.