The evaluative priming effect (i.e., faster target responses following evaluatively congruent compared with evaluatively incongruent primes) in nonevaluative priming tasks (such as naming or semantic categorization tasks) is considered important for the question of how evaluative connotations are represented in memory. However, the empirical evidence is rather ambiguous: Positive effects as well as null results and negatively signed effects have been found. We tested the assumption that different processes are responsible for these results. In particular, we argue that positive effects are due to target-encoding facilitation (caused by a congruent prime), while negative effects are due to prime-activation maintenance (caused by a congruent target) and subsequent response conflict. In 4 experiments, we used a negative prime-target stimulus-onset asynchrony (SOA) to minimize target-encoding facilitation and maximize prime maintenance. In a naming task (Experiment 1), we found a negatively signed evaluative priming effect if prime and target competed for naming responses. In a semantic categorization task (i.e., person vs. animal; Experiments 2 and 3), response conflicts between prime and target were significantly larger in case of evaluative congruence compared with incongruence. These results corroborate the theory that a prime has more potential to interfere with the target response if its activation is maintained by an evaluatively congruent target. Experiment 4a/b indicated valence specificity of the effect. Implications for the memory representation of valence are discussed. (Contains 6 tables, 2 figures and 9 footnotes.)