We analyzed the course of 186 patients with external gastrointestinal fistulas treated at the University of California Medical Center, San Francisco from 1968 to 1977. There were 82 patients in the earlier group (1968-1971) and 104 patients in the later group (1972-1977). The groups differed in that 35% of patients in the earlier group received TPN, but 71% of patients in the later group received TPN. Of the patients who did not receive TPN, 93% had been adequately nourished using tube feeding methods. The two groups were otherwise similar. The fistula-related mortality (11%) and the spontaneous closure rate of the fistulas (32%) was unchanged over the ten year period. Thus, the principal impact of TPN was to simplify the nutritional management rather than to alter the outcome. When malignancy, previous abdominal irradiation, Crohn's disease, or a short (<2 cm) fistula tract were present, spontaneous closure was less likely than when none of these factors were present (20% versus 47%). Sixty-eight per cent of the deaths occurred in patients with uncontrolled sepsis. Fifty per cent of the deaths were due to the primary disease and were unrelated to the fistula. Spontaneous closure could not be expected to start until sepsis was controlled. Because over 90% of patients whose fistulas closed spontaneously did so within one month after infection was eradicated, we recommend operative closure for most fistulas that persist beyond that time. The most reliable operation is excision of the bowel from which the fistula arises with end-to-end anastomosis. Fistulas not amenable to excision should be managed by bypass.