We have previously shown that patients with a positive family history of lung cancer did not exhibit characteristics expected if the familial association was a surrogate for a genetic mechanism which was absent in those without such a history. In this study, we examine the incidence of cancer (all sites) in two groups of families: (a) those ascertained through a patient with primary lung cancer (n = 359 families) and (b) through a community control subject (n = 234). The index subjects were excluded from all incidence data analyses. Most families (62 vs 57%) reported at least one member with a history of neoplastic disease. Patients' families were more likely to have 2 or more affected members (p less than 0.05), to exhibit a multigenerational pattern (33 vs 24%) and to have a higher risk of multiple tumours (OR = 1.5) in the same individual. However, most of the sites with the highest odds ratios (males, vocal cord/esophagus 8.3; colon/rectum 2.3; lung 1.96; females, pancreas 4.8; vocal/cord esophagus 4.1; lung 1.8) are smoking associated although these were not necessarily the most frequently affected sites. In conclusion, these data support an ecogenetic etiology of cancer within families.