Decompressive craniectomy in the treatment of severe traumatic brain injury (TBI) is controversial. We conducted a retrospective review of prospectively collected data on all patients requiring surgery for TBI from 1995 through 2001 at Cedars-Sinai Medical Center. Patients were separated into two groups: Group A, craniectomy, and Group B, craniotomy. We had 120 patients; 24 (20%) had craniectomy and 96 (80%) had craniotomy. There were no significant differences in demographics or Injury Severity Scores. The craniectomy group had significantly more TBI as evidenced by more frequently collapsed basilar cisterns on CT scan (P = 0.0001). There was no significant difference in actuarial survival between the groups: 52.8 per cent in the craniectomy group and 79.2 per cent in the craniotomy group (P = 0.08). Calculated mortality for craniectomy was 37.5 per cent versus 18.8 per cent for craniotomy (P = NS). We found four preoperative findings to be significant predictors of mortality: 1) Glasgow Coma Scale score, 2) Injury Severity Score, 3) Simplified Acute Physiology Score, and 4) Acute Physiology and Chronic Health Evaluation II. The type of surgery was not found to be a significant predictor of death even when adjusted for severity of injury. Craniectomy may be helpful for patients with TBI associated with preoperative CT scan evidence of basilar cistern collapse. This is evidenced by similar survival rates between the two groups despite clinical evidence of greater TBI among craniectomy patients.