In the same-different match task, masked priming is observed with the same responses but not different responses. Norris and Kinoshita's (2008) Bayesian reader account of masked priming explains this pattern based on the same principle as that explaining the absence of priming for nonwords in the lexical decision task. The pattern of priming follows from the way the model makes optimal decisions in the two tasks; priming does not depend on first activating the prime and then the target. An alternative explanation is in terms of a bias towards responding “same” that exactly counters the facilitatory effect of lexical access. The present study tested these two views by varying both the degree to which the prime predicts the response and the visibility of the prime. Unmasked primes produced effects expected from the view that priming is influenced by the degree to which the prime predicts the response. In contrast, with masked primes, the size of priming for the same response was completely unaffected by predictability. These results rule out response bias as an explanation of the absence of masked priming for different responses and, in turn, indicate that masked priming is not a consequence of automatic lexical access of the prime.