The steady-state {(1)H}-(15)N NOE experiment is used in most common NMR analyses of backbone dynamics to accurately ascertain the effects of the fast dynamic modes. We demonstrate here that, in its most common implementation, this experiment generates an incorrect steady state in the presence of CSA/dipole cross-correlated relaxation leading to large errors in the characterization of these high-frequency modes. This affects both the quantitative and qualitative interpretation of (15)N backbone relaxation in dynamic terms. We demonstrate further that minor changes in the experimental implementation effectively remove these errors and allow a more accurate interpretation of protein backbone dynamics.