Many months' treatment with daily oral mercaptopurine is an important part of therapy for nearly all children with lymphoblastic leukaemia (ALL). Even when prescribed the same dose based on body surface area, patients have widely different intracellular concentrations of drug metabolites. Whether this variation matters in terms of disease control is not yet clear. To find out, we followed up a large group of children with ALL in whom mercaptopurine-derived thioguanine nucleotides in the red cells were measured during treatment with an identical dose of mercaptopurine early in first remission. 172 unselected children (100 boys, 72 girls) were recruited between 1980 and 1992. At median follow-up of 5 years from diagnosis 42 (24%) had relapsed; 30 had erythrocyte thioguanine nucleotide concentrations below the group median (284 pmol per 8 x 10(8) erythrocytes) and 12 had values above the median. The actuarial relapse-free survival at 5 years was 63% in the below-median group and 84% in the above-median group (difference 21% [95% CI 3-39%], p = 0.0018). Multivariate analysis showed that erythrocyte thioguanine nucleotide concentration was independent of other prognostic variables including age, leukaemia immunophenotype, white-blood-cell count at diagnosis, trial protocol, and sex. Whatever the cause, in childhood ALL variable formation of intracellular mercaptopurine metabolites seems to be clinically important. Therapeutic schedules that include long-term daily oral mercaptopurine might be more effective if such metabolites are monitored.