In this paper it is demonstrated by simulation that, contrary to a widely held belief, pure seasonal mean shifts - i.e., seasonal structural breaks which affect only the deterministic seasonal cycle - really do matter for Dickey-Fuller long-run unit root tests.