Unconscious priming is sensitive to contextual factors. The present study examined this adaptive process using masked arrow primes (<< or >>). Some targets required specific "fixed" left/right responses (<< or >>) and others required "free" left/right responses (<>). Different groups (n = 30 each) received responsecongruent primes (SOA = 75 msec.) on 0.2, 0.5, or 0.8 of the fixed-response trials. Fixed responses were facilitated by congruent primes and free responses were faster when congruent with the prime. Critically, these masked priming effects emerged only in the 0.8 group. The pattern of extant prime-proportion effects in this paradigm best supports an adaptive associative-strength account rather than memoryrecruitment or response-bias-suppression accounts.