The detection of residual leukemia cells in the bone marrow of patients during morphologic remission has been greatly facilitated by use of the polymerase chain reaction (PCR) to amplify leukemia-specific sequences. While the current PCR strategies for estimating the amount of residual leukemia claim a detection sensitivity of one leukemia cell amongst 10(5) or 10(6) normal cells, a rigorous assessment of the relative error associated with these techniques has not been presented. We have developed a method of estimating the amount of residual leukemia in remission marrows that is analogous to the limiting dilution assays used to determine the frequency of immunocompetent cells in a responder cell population. Using this method we measured the fraction of all-or-none (i.e. positive or negative) reactions of the PCR amplification of the leukemia-specific IgH gene rearrangement in replicate samples of serial dilutions of DNA obtained from diagnostic bone marrow specimens from 15 children with B-precursor acute lymphoblastic leukemia (ALL). A sigmoid curve representing the fraction of positive PCR reactions at a given dilution of leukemia DNA was found to be the best fit to the data. The narrowness of the log-linear region of this curve prevents the direct application of the analysis methodology that has previously been described for limiting dilution assays. However, the residual leukemia burden during morphological remission in these 15 patients and in two additional patients who experienced relapse could be estimated by the described dilution analysis method using the best-fit equation. Furthermore, the data generated for diagnostic, remission and relapse marrow samples exhibited a small interspecimen variation. The results suggest that this method can reliably estimate residual leukemia over a range of five orders of magnitude. Although the PCR reaction appears to be one of the most sensitive methods for detecting residual leukemia, all techniques based on this procedure, including our own, must exhibit limitations inherent to the amplification process. Our estimates or relative error suggest that a realistic limit for the PCR estimation of residual leukemia lies in the range of one leukemia cell per 10(5) normal cells. The suggested method is rapid, technically simple and relatively inexpensive. Furthermore, the principles that it is based upon can be applied to any PCR-based strategy.