Liking of a formerly neutral stimulus can change when it co-occurs with positive or negative stimuli, a phenomenon called evaluative conditioning (EC). Prior research suggests that EC depends on remembering information about previous stimulus co-occurrences when the neutral stimulus is evaluated. According to a recently proposed integrative episodic retrieval account of contingency learning (Rothermund et al., 2024), the temporal distance to the last occurrence of a stimulus determines which of the previous experiences with the stimulus is retrieved from memory, favoring the most recent experience after short temporal delays and most frequent experiences after longer delays. In this research, we tested whether EC follows the same retrieval principles. Across three online experiments, we found that evaluations of a formerly neutral stimulus reflected the valence most recently paired with it when measured shortly after the pairing but reflected the most frequently paired valence when measured after a longer delay. Our research supports the role of episodic retrieval processes in EC and contributes to recent theoretical focus on memory mechanisms in EC. Furthermore, our findings highlight parallels to contingency learning, suggesting that episodic memory processes may govern various types of learning resulting from stimulus contingencies.