In a cohort of 593 long-term survivors of acute lymphocytic leukemia identified through the Children's Cancer Group (CCG), treatment abstracts were obtained and compared to protocol information on radiation therapy and intravenous chemotherapy. This was done in order to evaluate the actual compliance to protocol-specified treatment, and assess if protocol-specified doses can be used in studies of late effects of treatment. The compliance to protocol-specified type of treatment ranged between 95.3% (intrathecal methotrexate) and 98.6% (adriamycin) for chemotherapy, and between 94.1% (cranial radiation) and 97.0% (extended field radiation) for radiation. Concordance with the protocol-specified chemotherapy dose (+/- 25%) was 57.5% for adriamycin, 91.3% for daunomycin, and 48.5% for cyclophosphamide. When concordance was low, most patients received doses that were lower than expected. Concordance with chemotherapy was significantly lower for high-dose regimens than for low-dose regimens. Concordance with protocol-specified radiation dose (+/- 10%) was 87.4% for cranial radiation, 87.8% for spinal radiation, and 85.7% for extended field radiation. Concordance with treatment did not differ by gender, relapse status, or age at diagnosis. In this cohort of leukemia survivors, the validity of type of treatment was greater than the validity of dosage. Great care should be used when drawing conclusions about effects of treatment dosage. Although costly and time consuming, it appears that chart reviews are the most appropriate way to collect information about dose-related effects of therapy.