To compare three different approaches to the measurement of patients' expectations for care, we conducted a randomized controlled trial. Medical outpatients (n = 318) of a small (six-physician), single-specialty (internal medicine), academically affiliated private practice in Sacramento, California, were contacted by telephone the night before a scheduled office visit and enrolled over a 5-month period in early 1994. Patients were randomly assigned to receive: (1) a self-administered, structured, previsit questionnaire combined with a postvisit questionnaire; (2) an interviewer-administered, semistructured, previsit interview combined with a postvisit questionnaire; or (3) a postvisit questionnaire only. We assessed the number and content of patients' expectations by previsit questionnaire versus interview; the interaction between sociodemographic characteristics and survey method in predicting number of reported expectations; the effect of unfulfilled expectations elicited by questionnaire and interview on visit satisfaction; and the effect of unfulfilled expectations elicited directly and indirectly on visit satisfaction. Patients reported more expectations by structured questionnaire than semistructured interview (median 12 vs 3, p = .0001). Although there was no main effect of sociodemographic characteristics on expectations, nonwhite patients reported more expectations than white patients by questionnaire and fewer by interview. The number of interventions desired before the visit but not received (indirectly reported unfulfilled expectations) was associated with lower visit satisfaction regardless of whether a questionnaire or interview was used to elicit previsit expectations (p value for the interaction between number of expectations and survey method, > .20). Having more indirectly reported unfulfilled expectations was significantly associated with lower visit satisfaction even after controlling for the number of directly reported unfulfilled expectations (p = .021), but the incremental change in classification accuracy was small (increase in receiver-operating characteristic curve area, 3%). In conclusion, studies of patients' expectations for care must content with a substantial "method effect." In this study from a single group practice, patients checked off more expectations using a structured questionnaire than they disclosed in a semistructured interview, but both formats predicted visit satisfaction. Asking patients about interventions received in relation to their previsit expectations added little to simply asking them directly about omitted care. The interaction of survey method with ethnicity and other sociodemographic characteristics requires further study.