Four experiments in human predictive learning evaluated whether the extinction makes the acquisition context specific (EMACS) effect is attenuated when the increase in prediction error that extinction produces disappears. Participants had to evaluate the relationship between a given food (cue) that was ingested by an imaginary client of a given restaurant (context) and a potential gastric illness (outcome). The task was implemented using Gorilla online software. All participants received the relevant training in context A, and equivalent exposure to context B. Cue E was presented paired with the outcome in all groups. Cue E was then either extinguished (group E) or not extinguished (group NE), either previously or concurrently to training of the target cue (P). P was then tested in contexts A and B. When extinction was conducted concurrently, performance to P became context-dependent regardless of the number of extinction trials (12 or 24)-the EMACS effect. The EMACS effect disappeared when extinction was elongated to 24 trials, and conducted before acquisition of P. Implications of these results for attentional explanations of context processing are discussed. (PsycInfo Database Record (c) 2024 APA, all rights reserved).