A semantic priming task in which both pictures and words appeared as primes and targets was used to investigated the order of access to different kinds of stored information by these two types of representation. Level of prime processing was controlled by instructions to categorize (Experiment 1), name (Experiment 2), or report the color of the prime (Experiment 3). Similarly, target processing level was controlled in all three experiments by instructions to either name or categorize the target. Four major findings emerged. First, priming effects were independent of prime and target modality. Second, categorized targets benefitted from the prior presentation of a same category prime while named targets essentially did not. Third, categorized primes produced the greatest amount of priming while color-report primes produced the least. Fourth, both words and pictures could be named more rapidly than they could be categorized. Taken together the results of these experiments suggest that naming represents a shallower level of processing than categorization for both words and pictures. Implications of these results for current models of semantic priming are discussed.