The production effect is defined as better memory for items that were read aloud compared with items that were read silently. Quinlan and Taylor (2013) expanded the findings of the production effect by demonstrating that singing items produces even better recognition performance than reading aloud, and argued that this was due to enhanced relative distinctiveness. The current study tested three alternative accounts. In Experiment 1, we explored whether singing results in a larger production effect because it is deemed more bizarre than reading aloud. To address this, we tested a sample for whom singing does not seem bizarre: experienced singers. They also showed better recognition of items that were sung compared with those that were read aloud. In Experiment 2, we determined that singing appears to take longer than either reading aloud or reading silently; however, the possible effect of production time was further explored in Experiment 3. We did this by instructing participants to sing quickly, read aloud slowly, or read silently. Altering relative production times resulted in no discernible changes in subsequent recognition performance. Finally, in Experiment 4, we explored whether singing might strengthen the memory trace relative to reading aloud. We tested this by manipulating the production instruction between subjects. This eliminated the recognition advantage for both reading items aloud as well as for singing them aloud. Having ruled out these alternatives, we argue that singing improves subsequent recognition because it offers more distinctive elements than either reading aloud or reading silently. (PsycINFO Database Record (c) 2019 APA, all rights reserved).