The purpose of this study was to compare the accuracy of an actuarial method of predicting patients' preferences for life-sustaining treatment with the accuracy of surrogate decision makers. 401 outpatients 65 years old or older (mean = 73 years) and their self-designated surrogate decision makers recorded preferences for four life-sustaining medical treatments in nine hypothetical illness scenarios. The surrogates did not predict the patients' preferences more accurately than did an actuarial model using modal preferences. Surrogates' accuracy was not influenced by the use of an advance directive (AD) or discussion of life-sustaining treatment choices. In clinical practice, an actuarial model could assist surrogate decision makers when a patient has no AD, an AD is unavailable, a patient's AD is vague or describes treatment choices for only extreme or unlikely disease states, no proxy decision maker has been designated, or a patient was never competent.