The learned predictiveness effect or LPE is the finding that when people learn that certain cues are reliable predictors of an outcome in an initial stage of training (phase 1), they exhibit a learning bias in favor of these cues in a subsequent training involving new outcomes (phase 2) despite all cues being equally reliable in phase 2. In Experiment 1, we replicate the basic effect and found that the addition of a secondary memory task during phase 2 had no reliable influence on the LPE. In Experiment 2, we demonstrated that the same secondary task can either facilitate or disrupt the LPE, depending on whether the outcomes of phase 1 were affectively congruent or incongruent with the outcomes of phase 2. These findings are discussed in relationship to associative and inferential accounts of LPE.