Blink startle magnitude is linearly modulated by affect such that, relative to neutral stimuli, startle magnitude is inhibited during pleasant stimuli and potentiated during unpleasant stimuli. Andreatta, Mühlberger, Yarali, Gerber, and Pauli (2010), however, report a dissociation between startle modulation and explicit valence evaluations during backward conditioning, a procedure in which the unconditional stimulus precedes the conditional stimulus (CS). Relative to controls, startles elicited during the CS were inhibited, suggesting that the CS had acquired positive valence, but participants still evaluated the CS as unpleasant after the experiment. In Experiment 1, we aimed to replicate this dissociation using a trial‐by‐trial measure of CS valence to measure startle modulation and CS valence simultaneously during forward and backward differential fear conditioning. In Experiment 2, we examined whether early and late portions of the CS could acquire differential valence by presenting startle probes at early and late probe positions during the CS. In both experiments, the dissociation between startle modulation and explicit valence evaluations in backward conditioning replicated, with CS+ evaluated as less pleasant than CS‐, but startles elicited during CS+ inhibited relative to CS‐. In Experiment 2, we provide preliminary evidence that this inhibition was present early, but not late, during the CS+. The results replicate the dissociation between implicit and explicit CS valence reported by Andreatta et al. (2010) using a trial‐by‐trial measure of valence. We also provide preliminary evidence that this dissociation may occur because the implicit and explicit measures are recorded at different times during the CS presentation.