Three experiments examined why pictures produce priming on the word-fragment completion test, despite the fact that there is no match between the physical features of the picture and the word fragment. Pictures and words were presented as primes, and performance on the wordfragment completion test was measured; encoding and retrieval conditions were varied. Experiments 1 and 2 examined the role of picture labeling by increasing the presentation rate and by introducing a shadowing task during encoding; labeling appears to playa role in priming. In Experiment 3, the word fragments were presented for 500 msec, and subjects were required to provide a solution immediately. Word priming was unaffected, but picture priming was eliminated, suggesting that word fragments enable efficient recovery of perceptually similar primes (i.e., words), but slower and less direct recovery of conceptually similar but physically dissimilar primes (i.e., pictures).