The acetylator phenotype and genotype of AIDS patients, with and without an acute illness, was compared with that of healthy control subjects (30 per group). Two probe drugs, caffeine and dapsone, were used to determine the phenotype in the acutely ill cohort. Polymerase chain reaction amplification and restriction fragment length polymorphism analysis served to distinguish between the 26 known NAT2 alleles and the 21 most common NAT1 alleles. The distribution (%) of slow:rapid acetylator phenotype seen among acutely ill AIDS patients differed with the probe substrate used: 70:30 with caffeine versus 53:47 with dapsone. Phenotype assignment differed considerably between the two methods and there were numerous discrepancies between phenotype and genotype. The NAT2 genotype distribution was 45:55 slow:rapid. Control subjects, phenotyped only with caffeine, were 67:33 slow:rapid versus 60:40 genotypically. Stable AIDS patients, phenotyped only with dapsone, were 55:45 slow:rapid versus 46:54 genotypically. Following resolution of their acute infections, 12 of the acutely ill subjects were rephenotyped with dapsone. Phenotype assignment remained unchanged in all cases. The distribution of NAT1 alleles was similar in all three groups. It is evident from the amount of discordance between caffeine phenotype and dapsone phenotype or genotype that caution should be exercised in the use of caffeine as a probe for NAT2 in acutely ill patients. It is also clear that meaningful study of the acetylation polymorphism requires both phenotypic and genotypic data.