Relapse following remission induction chemotherapy remains a barrier to survival in approximately 20% of children suffering from acute lymphoblastic leukemia (ALL). To investigate the mechanism of relapse, 27 matched diagnosis and relapse ALL samples were analyzed for clonal populations using polymerase chain reaction (PCR)-based detection of multiple antigen receptor gene rearrangements. These clonal markers revealed the emergence of apparently new populations at relapse in 13 patients. More sensitive clone-specific PCR revealed that, in 8 cases, these "relapse clones" were present at diagnosis and a significant relationship existed between presence of the relapse clone at diagnosis and time to first relapse (P < .007). Furthermore, in cases where the relapse clone could be quantified, time to first relapse was dependent on the amount of the relapse clone at diagnosis (r = -0.84; P = .018). This observation, together with demonstrated differential chemosensitivity between subclones at diagnosis, argues against therapy-induced acquired resistance as the mechanism of relapse in the informative patients. Instead these data indicate that relapse in ALL patients may commonly involve selection of a minor intrinsically resistant subclone that is undetectable by routine PCR-based methods. Relapse prediction may be improved with strategies to detect minor potentially resistant subclones early during treatment, hence allowing intensification of therapy.