Perceptual learning improves with most basic stimuli. Interestingly, performance does not improve when stimuli of two types are randomly presented during training (roving). For example, there is no perceptual learning when left or right bisection stimuli with outer line distances of 20’ and 30’ are presented randomly interleaved from trial to trial. How can roving be explained? Perceptual learning is reward-based learning. A recent mathematical analysis showed that any reward-based learning system suffers from synaptic drift, which makes learning impossible, when two tasks are learned and the mean rewards of the tasks are not identical. Hence, we propose that perceptual learning fails in roving conditions because of the different rewards for the two roved tasks. The unsupervised bias hypothesis makes the surprising prediction that perceptual learning should also fail when an easy and a hard task are roved because of their different rewards. To test this prediction, we presented bisection stimuli with outer-line-distances of either 20’ or 30’. In both tasks, observers judged whether the central vertical line was closer to the left- or right-outer line. Task difficulty was adjusted by manipulating the center line’s offset. Easy and difficult discriminations corresponded to 70 and 87 percent correct respectively. As predicted, subjects failed to learn in this roving task for both bisection-stimulus types. Hence, an easy undemanding task can block perceptual learning of another task.