Mackintosh and his collaborators put forward an account of perceptual learning effects based, in part, on learned changes in stimulus salience. In the workshop held to mark Mackintosh's retirement, and published as a special issue of this journal, Hall discussed Mackintosh's theory and proposed his own alternative account. We now want to take the story forward in the light of findings and theoretical perspectives that have emerged since then. Specifically, we will argue that neither Mackintosh nor Hall was correct in his account of the principles that govern how changes in salience occur. Both supposed (in different ways) that such changes depend on the way in which the stimulus (or stimulus element) is predicted by another event. In contrast, theories of attentional learning have stressed the notion that changes in the properties of a stimulus might depend on the way in which it predicts its consequences. These theories have been concerned with attention-for-learning (associability). We now consider how the general principle they both employ might be relevant to the other forms of attention (for perception and for performance) that are, we will argue, critical for the perceptual learning effect.