It is a well known fact that stationarity and unit roots tests are seriously distorted, when the true data generating process is stationary around a broken trend. In this paper, the behaviour of Breitung's (J. Econom. (2002) 343-363) variance ratio test for unit roots is analyzed. It is shown that the test may be inconsistent against stationary alternatives with structural breaks. In addition, a new test procedure to account for structural breaks is proposed.