SUMMARYWe analyzed Kidney Transplant Registry survival data on transplants performed before 1 January 1967. Most transplants were performed irrespective of HL-A matching and were essentially random in this regard. Semilog plots of monozygous twin (MONO), cadaver (CAD), and living unrelated donor (LUD) transplant survival data were found to conform to a linear survivorship model (model 1). Sibling (SIB) and parent-child (PAR) data conformed to a convex survivorship model (model 2). These facts suggested that subpopulations of strikingly different immunological risk existed within SIB and PAR groups, but not within MONO, CAD, or LUD groups. Graphical procedures for estimating parameters of the survivorship models are described and, using these parameters, one can partition the probability of transplant failure into four clinically useful categories: acute technical risk, acute immunological risk, chronic aspecific risk, and chronic immunological risk. This breakdown of risk factors demonstrated that the total probabilities of failure from an immunological cause were 0.00, 0.41, 0.45, 0.89, and 0.93 for MONO, SIB, PAR, CAD, and LUD groups, respectively. The probabilities of immunological rejection thus stand in approximately the same proportions as the average number of HL-A mismatches for these five groups, namely, 0, 2, 2, 4, and 4. These results encourage one to believe that a successful matching program can significantly reduce CAD immunological risk. Moreover, if matching is successful, future CAD survival data will conform to a convex model 2 locus. Other points of biological interest are mentioned, and it is suggested that the partitioning of risk described in this paper can improve the evaluation of new therapies and matching procedures.