People normally change their attitudes towards stimuli in line with the valence of spatio-temporally co-occurring other stimuli; a phenomenon termed the evaluative conditioning effect. However, this attitude change is far from always valid, endorsed by, or in the best interest of the person as such effects are present in many circumstances (e.g., advertising, persuasion, etc.) where one might wish to prevent or reverse this influence. We investigate for the first time how people control the effects of evaluative conditioning by assessing the contents of the acquired representations. We distinguish between two general types of representations which are closely linked to two strategies of exerting control. We posit that people can either integrate or add the validity information to stored stimulus valence representations. Several indicators of these representations are derived from evaluative ratings, memory measures, and process dissociation parameters, and cross-validated in a set of three experiments. Additionally, the present research identifies a moderator of the control strategies that people use and the resulting representations. Specifically, we posit and find that people can integrate validity information into the stored valence representations mostly when they attempt to control the evaluative conditioning effect at the time of exposure to the evaluative information. Control attempts taking place later are more likely to add than to integrate the validity information. Our findings have implications for theories of evaluative conditioning, for the methodological approach to the study of controllable and uncontrollable processes, as well as for the protection from unwanted or unethical influences on attitudes.