This study examined witnesses' answers to repeated questions about a novel event, both within and across interviews. Ss in 4 age groups (4-, 6-, and 8-year-olds and adults; N = 133) individually witnessed an ambiguous incident. Some Ss were interviewed immediately and 1 week later; others were interviewed only once, 1 week later. Children were as accurate as adults when responding to open-ended questions, but 4-year-olds were more likely to change responses to yes-no questions. Adults speculated more frequently than children on a specific question about which they had no information, and answers to this question became more certain with repetition. An "inoculation" procedure was successful in reducing the frequency of inappropriate speculation. When openended questions were used, a moderate amount of repetition primarily influenced presentation style rather than accuracy.