This article presents a meta-analysis of research on evaluative conditioning (EC), defined as a change in the liking of a stimulus (conditioned stimulus; CS) that results from pairing that stimulus with other positive or negative stimuli (unconditioned stimulus; US). Across a total of 214 studies included in the main sample, the mean EC effect was d = .52, with a 95% confidence interval of .466-.582. As estimated from a random-effects model, about 70% of the variance in effect sizes were attributable to true systematic variation rather than sampling error. Moderator analyses were conducted to partially explain this variation, both as a function of concrete aspects of the procedural implementation and as a function of the abstract aspects of the relation between CS and US. Among a range of other findings, EC effects were stronger for high than for low contingency awareness, for supraliminal than for subliminal US presentation, for postacquisition than for postextinction effects, and for self-report than for implicit measures. These findings are discussed with regard to the procedural boundary conditions of EC and theoretical accounts about the mental processes underlying EC.