In two experiments, participants were exposed to a listing of actions performed by a fictitious Mr. X, over three days of his life. For most of his actions an outcome was described, but some were not followed by any outcome. On Day 3, Mr. X performed an action (the target action) that was followed by a novel outcome. For participants in the control condition, the target action that preceded the appearance of this outcome was also novel; for participants in the latent inhibition (LI) condition, Mr. X had performed the target action on repeated occasions during Days 2 and 3, without it producing any outcome. All the participants were tested on their ability to retrieve the action performed by Mr. X prior to the target outcome. In Experiment 1, retrieval of the target action (indicating a less effective target action–outcome association) was poorer in the LI than in the control condition. In Experiment 2, reducing the proportion (the density) of nontarget actions that brought outcomes during initial training was found to reduce the size of the LI effect. These results are predicted by the account of LI put forward previously [Hall, G., & Rodríguez, G. (2010). Associative and nonassociative processes in latent inhibition: An elaboration of the Pearce-Hall model. In R. E. Lubow & I. Weiner (Eds.), Latent inhibition: Data, theories, and applications to schizophrenia (pp. 114–136). Cambridge, England: Cambridge University Press]. A high density of predictive relationships ensures strong activation of the expectancy that some outcome will occur when the target action is first presented; this facilitates the formation of a target action–no-event association during training in the LI condition, thus enhancing the LI effect.