A study investigated translation problems arising in physician-patient interviews conducted in two languages with the help of an interpreter. Subjects were four adult native speakers of Gujarati, aged 42-70, whose physician interviews were videotape-recorded and translated, and the discourse was analyzed. Patients spoke in Gujarati, and the interpreters were family members. Questions were classified as simple, complex, or serial, and the frequency of each type was tabulated. Mistranslations in five categories (anatomical, symptoms and conditions, other terminology, euphemisms, loan words) were noted, and social and cultural factors evident in the exchanges were examined. It is concluded that while the interviews seem, on the surface and to the doctor, to be quite normal, they were often severely dysfunctional for linguistic and cultural reasons. The quality of information exchanged was such as to make correct initial diagnosis unlikely. Further study and consideration of the use of professional interpreters in such exchanges are recommended. (MSE)