On the basis of an extensive recursive partitioning analysis of 668 patients with newly diagnosed neuroblastoma registered on Pediatric Oncology Group (POG) studies between October 1981 and May 1987, four major subsets of patients were created. Important prognostic factors included the patient's stage of disease, age, and level of serum lactate dehydrogenase (LDH). After adjusting for these factors, no other clinical prognostic factors were significant. The implications for protocol design are that (a) fine tuning of current therapy should be sought for the two favorable disease patient subsets, while (b) novel aggressive therapies are needed for the two unfavorable disease patient subsets where the overwhelming majority are dying. This article may serve as a model for others investigating prognostic factors. The data were divided into two subsets: one was used for an exploratory analysis; the other was used to confirm the exploratory findings. Despite spite the large number of statistical tests performed, the likelihood that the findings can be attributed to chance can be dismissed as virtually zero.