It is widely believed that the prevalence of smoking among hospital patients is greater than that of the general population because many conditions for which patients are hospitalized are caused by or associated with smoking, and that this increased prevalence may bias results of case-control studies of tobacco-related diseases. For this reason, many authors have suggested excluding from the control series patients hospitalized for tobacco-related illnesses. The present study investigated potential selection bias for hospital compared to neighborhood controls in studying tobacco-related diseases. The 709 cases from six U.S. cities had tobacco-related cancers or myocardial infarction. They were individually matched to one hospital control and to one neighbor. After excluding patients with tobacco-related diseases, hospital controls were less often current smokers and more often former smokers than neighborhood controls. Among male ever smokers, hospital controls tended to smoke more cigarettes per day than neighborhood controls. Compared with the U.S. population, there was an overrepresentation of smokers in neighborhood controls rather than an under-representation of smokers in hospital controls. Relative risk estimates varied according to type of control. Choosing between hospital and neighborhood controls in case-control studies should be dictated by criteria related to the study hypothesis, participation, or cost. In particular, exclusion of hospital controls with diseases known to be tobacco-related seems to be a successful strategy for reducing selection bias.