Summary.52 fifth‐ and sixth‐grade children read a 900‐word text which was followed by a multiple‐choice test consisting of fact and inference questions. One and a half hours later, they were given feedback either with or without the original text present. Four days later all children were retested. On both the immediate and delayed test sixth‐graders performed better than fifth‐graders, while, especially in grade 5, more fact than inference questions were answered correctly. These results were interpreted as corroborating recent findings on developmental improvement in constructive cognition. Feedback on inference questions seemed somewhat more effective with the original text than without. The children correctly identified almost 90 per cent of their first‐test responses at the time of the delayed test. The identification patterns were interpreted as providing evidence against the interference‐perseveration explanation of the delay