Between January 1970 and August 1986, 3051 total hip arthroplasties (THAs) were performed at the authors' institution. Forty-seven hips in 43 patients developed deep infection. Using clinical and laboratory criteria, patients were classified into four modes of infection: (1) surgical contamination, which included hips (N = 13) with a suspicious clinical course (e.g., persistent elevation of sedimentation rate, early radiographic signs of loosening) without another identifiable source; (2) hematogenous spread, which included septic hips (N = 19) with a temporally related infectious source and an organism consistent with that source; (3) recurrent sepsis, septic failure in a previously infected hip (N = 13); and finally, (4) infection from direct or contiguous spread, which occurred in an additional two patients. Thus the majority of infections were nonsurgical. Patients with surgically acquired infections tended to present earlier, were less likely to require a Girdlestone procedure for salvage, and were more likely to be reimplanted (p = 0.024). The incidence of surgical infections decreased with the use of improved antiseptic techniques. The incidence of hematogenous infection, however, increased during the time that a cohort was followed. Additionally, in a previously septic hip, septic failure may occur as late as seven years after THA.