We have investigated whether the extent of residual leukemia at bone marrow harvest can predict subsequent relapse after autologous bone marrow transplantation (BMT). A total of 29 pre- and post-purged marrow samples from 15 patients with high-risk common acute lymphoblastic leukemia were examined. An accurate quantitation of residual disease was achieved by phage library assay using polymerase chain reaction to amplify the third complementarity determining region of the immunoglobulin gene. The estimated rate of disease-free survival at 3 years was significantly higher for the patients with less than 5% residual disease among total B cells than for those with greater than 5% before purging (87.5% vs 0%, P = 0. 0013). Furthermore, among patients with subsequent relapse, there was a linear correlation between the quantitated residual tumor burden of pre-purged marrow and remission duration after BMT (r2 = 0. 888). An accurate quantitative assessment of residual disease in the autograft has a high predictive value for subsequent relapse. A serial assay of residual disease would help us to individualize the treatment for each patient after induction or consolidation therapy.