Records of hospital inpatients were abstracted for 5,000 newly diagnosed cancer patients admitted in 1982-83 to 17 Comprehensive Cancer Centers and 17 Community Hospital Oncology Programs. Generally available data items (silent record rate less than 5 per cent for the typical institution) included: age, race, sex, dates of hospitalization, zip code of residence, pathological stage, dates of biopsy and surgery, numbers of nodes examined and positive, certain diagnostic procedures, and some radiotherapy descriptors. For other data items, there was enormous variability in completeness and high institution-to-institution variation. Record completeness did not differ consistently between comprehensive and community cancer centers. We conclude that the hospital patient record is useful for tracking the frequency of surgical and related events. However, studies of diagnostic and therapeutic procedures should not rely solely on the hospital medical record due to the high rates of silent records.