Many researchers regard the word-length effect (WLE) as one of the strongest pieces of evidence for time-based decay in short-term memory. We argue that the WLE is, in fact, undiagnostic for the decay hypothesis for two reasons. First, the WLE represents a correlation across words between articulation duration and memory performance, and articulation duration is inevitably confounded with other word characteristics. Recent research has confirmed that such confounds are responsible for much, maybe all, of the WLE. Second, there is strong evidence for an attentional mechanism of refreshing memory traces that can operate concurrently with articulation. Any viable decay-based model must include such a mechanism, but such a model no longer necessarily predicts a WLE, because longer spoken duration does not imply longer postponement of refreshing. We conclude that the WLE is not diagnostic for decay in short-term memory.