The determination of the resolution of cosmological N-body simulations, i.e., the range of scales in which quantities measured in them represent accurately the continuum limit, is an important open question. We address it here using scale-free models, for which self-similarity provides a powerful tool to control resolution. Such models also provide a robust testing ground for the so-called stable clustering approximation, which gives simple predictions for them. Studying large N-body simulations of such models with different force smoothing, we find that these two issues are in fact very closely related: our conclusion is that resolution in the non-linear regime extends, in practice, down to the scale at which stable clustering breaks down. Physically the association of the two scales is in fact simple to understand: stable clustering fails to be a good approximation when there are strong interactions of structures (in particular merging) and it is precisely such non-linear processes which are sensitive to fluctuations at the smaller scales affected by discretisation. Resolution may be further degraded if the short distance gravitational smoothing scale is larger than the scale to which stable clustering can propagate. We examine in detail the very different conclusions of a study by Smith et al. (2003), and find that the strong deviations from stable clustering reported by this work are the results of over-optimistic assumptions about resolution of the measured power spectra, and the reliance on Fourier space analysis. We emphasize the much poorer resolution obtained with the power spectrum compared to the two point correlation function.