A protocol for selective ordering of 12 preoperative tests, according to clinical status and type of surgery, was prospectively tested for one year in a teaching hospital. 3866 consecutive surgical patients had an average of about 4 tests each. The possible value of tests that were omitted was assessed in the light of events during and after operation. According to predetermined criteria, 0.4% of non-ordered tests would have been potentially useful; but in the opinion of the anaesthetists, only 0.2% would actually have been useful. The protocol therefore had little adverse effect on patient care and was acceptable to clinicians.