After several preliminary joint sessions a doctor and a nurse independently constructed summary cards for 100 alphabetically consecutive case records and coded them according to the classification of the Royal College of General Practitioners. The searches and codings were repeated after an interval of at least one month. The error rates of the doctor and the nurse were similar. The interobserver concordance for the ratings was acceptable for broad diagnostic categories, but was less good when strict criteria were applied. It is concluded that the task may be delegated.