Twenty patients with clinically definite, stable multiple sclerosis were examined independently by three of us at the same visit and given scores on the Ambulation Index, Expanded Disability Status Scale, and Kurtzke Functional System scales. Observer error accounted for 12% to 55% of the variation observed between individual Kurtzke Functional System scores, 17.1% of the variation observed between the patients' Expanded Disability Status Scale scores, and only 3.9% of the variation between Ambulation Index scores. The implications of these findings for the choice of scales in clinical trials are described.