We investigated the encoding mechanisms involved in the perceptual recongnition of words and pictures. Latencies in naming word and picture targets were analyzed as a function of several characteristics of a preceding prime, including whether it was a word or a picture, its duration of exposure, the interval between the prime and target onset, and whether or not the prime was consciously identified and reported by the subject. Results indicated that a common semantic code is available that can represent the meaning of either a word or a picture. This semantic representation, however, appears to be more easily activated by picture primes than by word targets. Despite the advantage for pictures with respect to semantic activation, overall proccessing in the naming task was slower and more attention demanding for pictures than for words. Comparison of our data with data on classification, in which an opposite pattern occurs (overall processing appears to be slower and more attention demanding for words than for pictures), suggests that, on the average, pictures have faster and more automatic access to their names but that words have faster and more automatic access to their names tant to their meanings. This conclusion concerning the relative ability of stimuli to activiate different kinds of internal representations has implications for a theory of the basis and development of automaticity.