Objective: This study aimed to determine the accuracy with which children aged 5 to 7 years were able to report the food eaten at a school lunch. Subjects/setting: Two hundred and three children (103 boys, 100 girls) aged 5-7 years were recruited from three primary schools in Oxford. Design: Trained investigators made observational records of the school dinner and packed lunch intakes of four or five children per session. Children were interviewed within two hours of finishing the lunchtime meal and asked to provide a free recall of their meal. When the child had completed the recall, non-directive prompts were used to assess if the child was able to remember anything else. Foods recalled were classified as matches (recalled food agreed with observation), omissions (failed to report a food observed) or phantoms (recalled food was not observed). Results: The percentage of accurate recall was significantly higher (P<0.01) in children eating packed lunch (mean 70+/-29%) than in children consuming school dinners (mean 58+/-27%). This difference may have been due to increased familiarity of foods in packed lunches. Leftovers were not readily reported in this age group. Prompts and cues enhanced recall by all children. Conclusions: This study indicated that there was a wide range in the ability of children aged 5-7 years to recall intake from a packed lunch and/or school dinner. This dietary assessment method is unlikely to be suitable at an individual level. Investigators using dietary recall to estimate food intake in children aged 5-7 years need to be aware of the limitations of this method.