Between 1981 and 1999, 890 Austrian children with acute lymphoblastic leukemia (ALL) were treated in 5 consecutive trials using protocols from the Berlin-Frankfurt-Münster (BFM) Group. In the trials BFM-A (Austria) 81 and ALL A 84, treatment stratification was performed using a risk factor, which was calculated from the initial peripheral blast cell count, and size of liver and spleen. In the following studies (BFM-A 86, 90 and 95) early response to a 7-day systemic mono-therapy with prednisone (as measured by the peripheral blast cell count) was used as an overriding stratification factor; in order to reduce the need for cranial radiotherapy, all patients received high-dose methotrexate (5 g/m2) for preventive central nervous system treatment. Event-free survival (EFS) rates increased from study BFM-A 81 (n = 141, probability (p) of EFS: 59% +/- 4%) and study ALL A 84 (n = 127, pEFS: 67% +/- 4%) to 77% +/- 4% in trial BFM-A 86 (n = 142), 79% +/- 3% in trial BFM-A 90 (n = 256), and 84% +/- 3% in trial BFM-A 95 (n = 224). However, the prognosis for high-risk patients has not significantly improved within the last 20 years (pEFS: 50%). Furthermore, conventional risk factors such as leukocyte count and age at time of diagnosis, could not be used to indicate patients in the low and intermediate risk group who might eventually relapse. Thus, in trial BFM-A 2000, detection of minimal residual disease by polymerase chain reaction-based methods after 5 and 12 weeks of therapy was introduced for treatment stratification. Minimal residual disease was prospectively shown to predict relapses more precisely and, as a matter of fact, may allow a more exact definition of which patients are at risk and which patients belong to the subgroup with a good prognosis despite reduced treatment.