Clinicians usually employ indirect measures of cognitive and physical function in order to assess medical decision-making capacity. We tested a reference group of well elderly (Mini-Mental State Exam [MMSE] score = 29.1 +/- 0.8, mean +/- SD), for their understanding of three increasingly complex, hypothetical treatment situations or "vignettes"--use of a hypnotic, need for thoracocentesis, and desire for CPR. From this, we have developed a more direct, Guttman-like assessment of decision-making capacity. Of 51 Veterans Affairs nursing home residents (MMSE score = 22.4 +/- 6.9), only 33.3% demonstrated intact decision-making capacity by this method, whereas 77% were felt by their primary physicians to be capable of giving consent for oral surgery; 37.3% had very impaired decision-making capacity; and 29.4% were intermediate in this ability. Judged against our more direct assessment of decision-making capacity, primary physicians' judgment of capacity for consent was 31% to 39% sensitive in identifying impaired decision-making and the MMSE was 53% to 63% sensitive. These measures were 100% and 82% to 83% specific in identifying intact decision-making capacity, respectively. We conclude that (1) more directly assessed decision-making capacity varies noticeably among elderly nursing home residents and correlates in only limited fashion with frequently used cognitive screening methods; and (2) cognitive screening tests underestimate the prevalence of impaired decision-making capacity in this population. For informed consent and advance directives, our study suggests that decision-making capacity should be directly, rather than indirectly, assessed.