Changing attitudes by repeated co-occurrences of initially neutral stimuli (conditioned stimuli [CSs]) with affective entities (unconditioned stimuli [USs]) is called evaluative conditioning (EC). The vast majority of EC procedures in the literature are “forward” in nature, presenting the CS before the US. Scant empirical research into the issue has argued that forward procedures are more effective than backward procedures, but this research suffers from methodological issues while a meta-analysis indicated no difference. Two experiments show that backward conditioning procedures are equally effective in changing attitudes as forward conditioning procedures. Memory measures show that memory associations are equally strong from the CSs to the USs as from the USs to the CSs, irrespective of the presentation order (forward vs. backward) of the stimuli. Together the data support the proposition that the associations generated by EC are symmetric and bidirectional, rather than unidirectional, in nature.