In the valence contingency learning task (VCT), participants evaluate target words which are preceded by nonwords. Nonwords are predictive for positive/negative evaluations. This produces (a) robust contingency learning effects, reflected in faster and more accurate performance for highly contingent nonword-valence pairings and (b) less robust evaluative conditioning (EC) effects, reflected in more positive ratings of nonwords that were highly predictive for positive (vs. negative) evaluative responses. In a highly powered (N=129) preregistered study, we investigated both effects and assessed whether they are a consequence of episodic retrieval of incidental stimulus-response (SR) episodes or contingency awareness. Participants were either explicitly instructed about contingencies (instructed learning group) or not (incidental learning group). Both groups then worked through the VCT, an explicit rating task, and a contingency awareness test. Both groups showed contingency learning effects and EC effects for nonwords. Hierarchical multi-level analyses showed that contingency learning effects were fully explained by episodic retrieval of SR episodes only in the incidental learning group, but not in the instructed learning group. EC effects in turn were fully explained by nonword-specific contingency awareness. Implications for theories on processes underlying contingency learning and evaluative conditioning are discussed.