Minimal residual disease (MRD) quantified after induction treatment of childhood acute lymphoblastic leukemia (ALL) predicts risk of relapse. It has been assumed that early relapses derive from a residual population of leukemic cells, which is still present after induction and that relapsed disease will consequently be more resistant to treatment. To test these hypotheses, we performed a prospective study on patients treated according to the frontline-trial ALL-BFM 2000, which used MRD response for risk-group stratification. Patients (n=45) showed a median time to relapse of 1.5 years. In 89% of patients at least one T-cell-receptor/immunoglobulin gene rearrangement chosen for initial MRD quantification remained stable; however, at least one of the preferred markers for MRD stratification at relapse was different to diagnosis in 50% of patients. A similar proportion of very early, early and late relapses appeared to gain a marker at relapse although backtracking-analysis revealed that in 77% of cases, the gained markers were present as small sub-clones at initial diagnosis. Comparing initial and relapse MRD response to induction, 38% of patients showed a similar, 38% a better and 25% a poorer response after relapse. These data demonstrate an unexpectedly high clonal heterogeneity among very early/early relapses and challenge some current assumptions about relapsed ALL.