In this paper, we show that spurious cointegration can occur when there are breaks in the variances of the innovation errors of time series, especially when the breaks occur early in the sample period. An empirical example is provided to demonstrate the case.