We revisit a previous study on inter-session variability (McGonigle et al. [2000]: Neuroimage 11:708-734), showing that contrary to one popular interpretation of the original article, inter-session variability is not necessarily high. We also highlight how evaluating variability based on thresholded single-session images alone can be misleading. Finally, we show that the use of different first-level preprocessing, time-series statistics, and registration analysis methodologies can give significantly different inter-session analysis results.