The association between social connections and cancer incidence, mortality, and prognosis during 17 years of follow-up was examined in a population-based sample of 6,848 adults who lived in Alameda County, California, in 1965. Estimates of relative hazards were derived from Cox regression modeling, adjusting for age, smoking, physical health at baseline, alcohol consumption, and adjusted household income. Women who were socially isolated were at significantly elevated risk of dying of cancer of all sites and of smoking-related cancers. Social connections were not prospectively associated with cancer incidence or mortality among men, but men with few social connections showed significantly poorer cancer survival rates. These patterns of risk are consistent with the biology of different cancer outcomes. They also suggest a different role for social isolation in cancer among men and women.