Hirshman and Lanning ([1999]) failed to replicate findings reported by Conway and Dewhurst ([1995]). There were, however, critical differences in design between the two sets of experiments. For example, Hirshman and Lanning used a within-subject design and a short retention interval, whereas Conway and Dewhurst used a between-subjects design and longer retention intervals. We demonstrate in a new series of experiments that the Conway and Dewhurst findings do replicate when the correct design is used and, moreover, we show that the design differences in Hirshman and Lanning's experiments account for their findings. Finally, we develop an account in terms of the self that can explain the complex pattern of findings, a pattern which lies beyound the scope of simple trace strength models of memory.