In novel situations, learning is biased towards information that has a degree of prior predictive utility. In human learning, this is termed the learned predictiveness effect and has proved critical in theorising about the role of attention in learning. Two experiments are reported in which the relative contribution of controlled and automatic processes to learned predictiveness are investigated. Experiment 1 showed that while learned predictiveness is susceptible to instructional manipulation, this effect is partial. Experiment 2 manipulated predictive utility and instruction orthogonally in order to test the potential involvement of automatic processes. It was found that even when cues were explicitly instructed as causal, learning was biased in favour of previously predictive over previously non-predictive cues. Interestingly, this was reversed for cues instructed as irrelevant. This suggests that learned predictiveness benefits attentional control, whereby information is both easier to attend and ignore.