Summary Detection of minimal residual disease (MRD) after induction and consolidation therapy is highly predictive of outcome for childhood acute lymphoblastic leukaemia (ALL) and is used to identify patients at high risk of relapse in several current clinical trials. To evaluate the prognostic significance of MRD at other treatment phases, MRD was measured by real‐time quantitative polymerase chain reaction on a selected group of 108 patients enrolled on the Australian and New Zealand Children’s Cancer Study Group Study VII including 36 patients with a bone marrow or central nervous system relapse and 72 matched patients in first remission. MRD was prognostic of outcome at all five treatment phases tested: at day 15 (MRD ≥ 5 × 10−2, log rank P < 0·0001), day 35 (≥1 × 10−2, P = 0·0001), 4 months (≥5 × 10−4, P < 0·0001), 12 months (MRD ≥ 1 × 10−4, P = 0·006) and 24 months (MRD ≥ 1 × 10−4, P < 0·0001). Day 15 was the best early MRD time‐point to differentiate between patients with high, intermediate and low risk of relapse. MRD testing at 12 and particularly at 24 months, detected molecular relapses in some patients up to 6 months before clinical relapse. This raised the question of whether a strategy of late monitoring and salvage therapy will improve outcome.