The value of cancer treatment was assessed using a 'natural experiment' where patients who refused treatment served as no-treatment controls in a situation where withholding treatment to form a control group is unethical. Each cancer patient who refused treatment in Alberta, Canada between 1975 and 1988 was compared with five subjects who accepted treatment, matched on cancer site, age, number of cancers, and time period. Variables associated with treatment-refusal were included in Cox's proportional hazards model of survival, with death from cancer as the endpoint and deaths from other causes as censored observations. Treatment was refused at a rate of 7.5 per 1,000. One-third of patients who refused treatment had lung cancer and most had unstaged disease. Treatment refusal was associated with a difference in median survival of approximately nine months. Site-specific analyses showed a range of effects. Case fatality among the treated patients fell by approximately 10 percent during the 14-year study period. Even in advanced disease, treatment can result in improved survival. However, the results of this study must be interpreted with caution and cannot be generalized to all cancer patients.