Instructions highlighting that backward conditional stimuli (CSs) stop unconditional stimuli (USs) result in their acquiring valence opposite to that of the US on explicit measures of valence. We assessed whether such instructions would influence startle blink modulation in the same way. Two groups were presented with concurrent forward and backward evaluative conditioning (CS-US-CS) using cartoon aliens as CSs, and pleasant, neutral, and unpleasant sounds as USs. Startle magnitude was measured during conditioning and valence ratings were assessed after conditioning. Participants in the "start-stop" instructions group (n = 41) were instructed to learn whether CSs started or stopped US presentations, while participants in the "observe" instructions group (n = 41) were told to pay attention to the stimuli as they would be asked questions about them after the experiment. In the start-stop instructions group backward CSs paired with positive USs were rated as less pleasant than backward CSs paired with neutral and negative USs (contrast effect), whereas ratings of backward CSs did not differ in the observe instructions group. Startle magnitude was larger during backward CSs paired with positive USs in comparison to CSs paired with neutral or negative USs in both instruction groups. Startle blink modulation was unaffected by instructions, suggesting that startle indexes the emotional state at the time of probe presentation rather than CS valence based on propositional information about the function of the CS.