Experiments 1-3 show that practice classifying visible words as pleasant or unpleasant enables word parts as small as single letters to function as subliminal primes having valence of words they came from. Subliminal primes were 33 ms in duration, forward and backward masked. Experiments 4 and 5 showed no subliminal priming by words that lacked parts of practiced words. Experiment 5 also included longer prime durations, at which unpracticed words DID produce priming. The latter effect was both unconscious and qualitatively different from the shorter-prime-duration effect, revealing a second mode of unconscious cognition..