The phonological similarity effect (or acoustic confusion effect) consists of poor serial recall performance for lists composed of similar-sounding words compared to lists of dissimilar-sounding items. Building on the work of Nairne and Kelley (1999), the extent to which this classical verbal short-term memory effect changes over the course of a retention interval was investigated. The impact of the phonological similarity on memory performance over time was compared in three paradigms: order reconstruction, serial recall, and free recall. Participants were presented with two blocks of 15 lists composed of five monosyllabic words, one block containing phonologically similar words and the other phonologically distinct words. Recall occurred either after 2, 8, or 24 seconds retention interval filled with a digit shadowing task. Our results confirmed and extended those of Nairne and Kelley. In order reconstruction and serial recall tasks, a classical phonological similarity effect (i.e., decrement in performance with similar lists) was observed for the 2-s retention interval. This effect disappeared after a brief period of digit shadowing (i.e., 8-s delay) and a beneficial effect of similarity was observed for the 24-s retention interval. In the free recall task, the phonological similarity produced better performance whatever the delay between presentation and recall. Those results are discussed in the light of Nairne's (1990a) feature model and confirm the role of item-specific and inter-item processes in order reconstruction, serial recall, and free recall.