Background. Registry data can provide valuable information about possible treatment effects; however, pretreatment differences in patient characteristics may influence treatment assignment. Careful analysis must therefore be undertaken when evaluating treatment differences in the context of nonrandomized studies so that the impact of treatment selection bias is minimized. Methods. A multivariable risk factor analysis of adult patients registered in the US Renal Data System who received a primary renal allograft during 1995 to 1998 was undertaken to compare 3-year graft survival using tacrolimus or Neoral with mycophenolate mofetil (MMF) and steroids. Results. In total, 9,449 patients were included (cadaveric donor n = 6,011; living donor n = 3,438). Patients (2,130) received tacrolimus, and 7,319 received Neoral. At 3 years posttransplant, the proportion of cadaveric donor recipients experiencing all causes of graft loss was 10.0% for tacrolimus and 10.6% for Neoral; for living donor recipients these figures were 6.5% and 6.7%, respectively (unadjusted Kaplan-Meier analysis). The incidence of graft failure excluding death was also similar between the two groups. With Cox proportional hazards modeling, the adjusted relative hazard of 3-year graft failure for cadaveric donor patients taking tacrolimus versus Neoral was 1.02 (95% confidence interval [CI] 0.8-1.3), and for living-donor recipients it was 1.15 (95% CI 0.8-1.8). Conclusions. These results indicate excellent 3-year graft survival for both cadaveric and living-donor renal-transplant patients receiving either Neoral or tacrolimus with MMF and steroids, with no significant differences between treatment groups. On the basis of these results, relative cost-effectiveness may become increasingly important in selection of tacrolimus or Neoral as primary immunosuppressant for renal-transplant patients.