The aim of this paper is to examine, using Monte Carlo experiments, the behaviour of unit root tests with a changing mean at an unknown date (Perron and Vogelsang, Journal of Business and Economic Statistics, 1992, 10, 301-320), in the presence of IGARCH errors. We find that the empirical sizes of the test statistics are significantly above the nominal ones and that distortions are considerably amplified by a conditional Student's t distribution instead of a normal one.