This paper describes several experiments that prove that standard motion correction methods may induce spurious activations in some motion-free fMRI studies. This artifact stems from the fact that activated areas behave like biasing outliers for the difference of square-based measures usually driving such registration methods. This effect is demonstrated first using a motion-free simulated time series including artificial activation-like signal changes. Several additional simulations explore the influence of activation amplitude and extent. The effect is finally highlighted on an actual time series obtained from a 3-T magnet. All the experiments are performed using four different realignment methods, which allows us to show that the problem may be overcome by methods based on a robust similarity measure like mutual information.