Does retrieval latency reflect variations in the strength of associations in episodic memory? In three experiments, subjects were given a single study and test trial on each of five lists of 10 paired associates. Spoken recall latencies were measured. When the subjects were later given a second test, initial recall latency was systematically related to intertest retention--that is, the faster the initial correct recall of a pair, the more likely a pair was to be recalled at the second test. This effect occurred at retention intervals of 5 min, 30 min, and 24 h and was present in the data for individual subjects. The results are consistent with the classical view of latency as a measure of trace strength and stand in sharp contrast with results reported by Benjamin, Bjork, and Schwartz (1998) that showed that fast retrievals from semantic memory were more poorly retained than slower ones.