We examined the outcome of the 489 children with acute lymphoblastic leukaemia (ALL) who relapsed in the UKALL X trial, and produced graphical displays of adjusted comparisons of event-free survival (EFS) for chemotherapy versus bone marrow transplantation (BMT) from a sibling or volunteer unrelated donor, and autologous BMT (ABMT). EFS at 5 years was only 3% (95% CI 0-6%) for children who relapsed in the bone marrow (BM) within 2 years of diagnosis, irrespective of type of post-relapse treatment, whereas for those with late extramedullary relapse it was 66% (95% CI 48-85%). Comparison of the types of treatment did not show benefit for ABMT. For allogeneic BMT the overall reduction in the odds of an event was 26% (95% CI 1-51%) (2P= 0.05), resulting in an absolute increase in 5-year event-free survival of 14% (from 26.4% to 40.7%). New approaches are needed for children with early BM relapses whose prognosis is virtually hopeless with current therapy; however, a conventional chemotherapy approach may be justifiable for late extramedullary relapses. For the remaining patients (71%), with later BM or early extramedullary relapses, the optimal treatment is still not clear. This uncertainty warrants a formal randomized comparison of BMT and chemotherapy, to avoid the biases due to unmeasurable selection factors.