Memory-derived predictions help us to anticipate incoming sensory evidence. A mismatch between prediction and evidence leads to a prediction error (PE). Previous research suggested that PEs enhance memory of the surprising events. Here, we systematically investigated the effect of PE on episodic memory in children (10-12 years old), younger adults (18-30 years old), and older adults (66-70 years old). Participants learned visual object pairs over 2 days. On Day 3, new objects were shown among the pairs, either after the first item of a pair (violation items), that is, instead of the second item, or between pairs (nonviolation items), that is, when no specific predictions were possible. Our results did not reveal a significant boosting effect of PE on memory in any of the age groups. In contrast, in children, violations resulted in lower memory specificity compared with nonviolations. Older adults showed lower memory specificity than the other age groups across violations and nonviolations. We conclude that the beneficial effect of PE on episodic memory may be less consistent than theoretically postulated and may not always be observed in experimental settings involving statistical learning and item-specific violations, and that children's memory specificity may even suffer from PE. (PsycInfo Database Record (c) 2025 APA, all rights reserved).