Adiabaticity plays a central role in modern magnetic resonance experiments, as excitations with adiabatic Hamiltonians allow precise control of the dynamics of the spin states during the course of an experiment. Surprisingly, many commonly used adiabatic processes in magnetic resonance perform well even though the adiabatic approximation does not appear to hold throughout the process. Here we show that this discrepancy can now be explained through the use of Berry's superadiabatic formalism, which provides a framework for including the finite duration of the process in the theoretical and numerical treatments. In this approach, a slow, but finite time-dependent Hamiltonian is iteratively transformed into time-dependent diagonal frames until the most accurate adiabatic approximation is obtained. In the case of magnetic resonance, the magnetization during an adiabatic process of finite duration is not locked to the effective Hamiltonian in the conventional adiabatic frame, but rather to an effective Hamiltonian in a superadiabatic frame. Only in the superadiabatic frame can the true validity of the adiabatic approximation be evaluated, as the inertial forces acting in this frame are the true cause for deviation from adiabaticity and loss of control during the process. Here we present a brief theoretical background of superadiabaticity and illustrate the concept in the context of magnetic resonance with commonly used shaped radio-frequency pulses.