The level of minimal residual disease (MRD) early in treatment of acute lymphoblastic leukemia (ALL) strongly predicts the risk of marrow relapse. As a variety of methods of varying complexity have been separately used for detecting and quantifying MRD, we compared the prognostic utility of three methods measurement of blast percentage on day 14 of treatment, detection of monoclonality on day 14 or day 35, and measurement of MRD by PCR-based limiting dilution analysis on day 14 or day 35. The study group comprised 38 children aged 1-15 with Philadelphia-negative B-lineage ALL who were uniformly treated and followed until relapse or for a minimum of 5 years. We also studied some of the technical factors which influence the ability to detect MRD. Measurement of blast percentage on day 14 by an expert morphologist, detection of monoclonality on day 35, and PCR-based measurement of MRD levels on days 14 and 35 all showed significant ability to divide patients into prognostic groups. Measurement of blast percentage on day 14 by routine morphology or detection of monoclonality on day 14 were not useful. The quality of DNA samples varied greatly, as determined by amplifiability in the PCR. However, virtually all amplifiable leukemic targets in a sample were detectable which suggests that the level of detection achieved by limiting dilution analysis is essentially determined by the amount of DNA which it is practicable to study. We conclude that quantification of MRD at the end of induction provides the full range of prognostic information for marrow relapse but is complex; detection of monoclonality on day 35 is simple and has good positive predictive value; and quantification of MRD on day 14 merits further study. PCR-based methods for measurement of MRD levels should incorporate a correction for variation in DNA amplifiability.