BACKGROUND: A study was conducted to test whether patient reports of medical errors via surveys could produce sufficiently accurate information to be used as a measure of patient safety. METHODS: A survey mailed regularly by a large multispecialty medical group to recent patients to assess their satisfaction and error experiences was expanded to collect more details about the patient-perceived errors. Following an initial mailing to 3,109 patients and parents of child patients soon after they had office visits in June 2005, usable mailed or phone follow-up responses were obtained from 1,998 respondents (65.1% adjusted). Responses were reviewed through a two-stage process that included chart audits and implicit physician reviewer judgments. The analysis categorized the review results and compared patient-reported errors with satisfaction. RESULTS: Of the 1,998 respondents, 219 (11.0%) reported 247 separate incidents, for a rate of 12.4 errors per 100 patients. After complete review, only 5 (2.0%) of these incidents were judged to be real clinician errors. Most appeared to represent misunderstandings or behavior/communication problems, but 15.4% lacked sufficient information to categorize. Women, Hispanics, and those aged 41-60 years were most likely to report errors. Those respondents making error reports were much more likely to report visit dissatisfaction than those not reporting them (odds ratio [OR] = 13.8, p < .001). DISCUSSION: Although patient reports of perceived errors might be useful to improve the patient experience of care, they cannot be used to measure technical medical errors and patient safety reliably without added evaluation. This study's findings need to be replicated elsewhere before generalizing from one metropolitan region and a patient population that is about two-thirds members of one health plan.