Are priming effects in implicit memory tasks produced by changes in lexical activation or by the formation of an episodic memory trace of the prime? If the latter view is correct, then no priming should be observed if subjects are unaware of the priming stimulus. Four experiments are reported in which the effect of a masked priming word upon subsequent performance on stem-and fragment-completion tasks was examined. Strong and consistent effects were observed in both tasks when there was a short delay (1 sec) between the presentation of the prime and the test item, but at longer delays (20 sec), these effects appear to have dissipated. These results are consistent with the lexical activation view. However, they are also compatible with the view that a masked prime generates a special type of memory trace that is inaccessible to conscious awareness and is extremely short-lived. It is argued that further evidence is required to decide between these alternatives.