Hospital morbidity data are useful for administrative purposes, but because Canadian data are based on the number of hospital visits for a given diagnosis rather than the number of patients with the diagnosis, they have historically been ineffective in determining how many people have been hospitalized with a given condition. Now, by linking computerized patient data, the records of the same patient can be combined. Linked records can be used to estimate disease prevalence, examine health care utilization, and evaluate the effectiveness of medical treatments, procedures and programs. Hospital morbidity records for the fiscal year 1989/90 were linked by person to study hospital utilization in New Brunswick and Saskatchewan. Approximately 11% and 12% of their populations, respectively, were admitted to hospital in the study period. The percentage of the population that was hospitalized increased with age from 50 onwards. About one in four hospital patients were admitted to hospital more than once in 1989/90, and approximately 4% were admitted four or more times. Cancer diagnoses were associated with the highest hospital re-admission rates. About half of hospitalized patients spent five days or less in hospital over the period studied. At the other extreme, 10% of patients, referred to as "high users," accounted for about half of the hospital days--but only 1% of the population in these two provinces. Typically, high users are patients with chronic conditions or illnesses severely affecting cognitive or physical abilities. This profile of high users suggests that high medical costs are due not so much to intensive care of terminally ill patients, but to ordinary medical and palliative care of chronically and seriously ill patients. Restructuring health care data so that all of the records for one patient could be linked would help identify problem areas in the health care system and help evaluate new ways of delivering health care.