The reliability of diagnoses of 46 child psychiatric inpatients made using three independent instruments--an unstructured interview with the parent and child that produced the chart diagnosis, a structured interview with the parent, and a structured interview with the child--was assessed. These diagnoses were then compared with a review diagnosis, which was based on all information available at the patient's discharge. Substantial disagreement between the diagnoses resulting from the structured and unstructured interviews was found in one-third of the cases. The review diagnosis disagreed with the chart diagnosis in 33 percent of the cases but disagreed with the diagnosis resulting from the structured interview with the parent in 13 percent of the cases. While structured interviews can contribute to the standardization of diagnoses, the reliability and validity of child psychiatric diagnoses remain problematic.