Estimates of the incidence of out-of-hospital primary cardiac arrest (CA) have typically relied solely upon emergency medical service or death certificate records and have not investigated incidence in clinical subgroups. Overall and temporal patterns of CA incidence were investigated in clinically defined groups using systematic methods to ascertain CA. Estimates of incidence were derived from a population-based case-control study in a large health plan from 1986 to 1994. Subjects were enrollees aged 50 to 79 years who had had CA (n = 1,275). A stratified random sample of enrollees who had not had CA was used to estimate the population at risk with various clinical characteristics (n = 2,323). Poisson's regression was used to estimate incidence overall and for 3-year time periods (1986 to 1988, 1989 to 1991, and 1992 to 1994). The overall CA incidence was 1.89/1,000 subject-years and varied up to 30-fold across clinical subgroups. For example, incidence was 5.98/1,000 subject-years in subjects with any clinically recognized heart disease compared with 0.82/1,000 subject-years in subjects without heart disease. In subgroups with heart disease, incidence was 13.69/1,000 subject-years in subjects with prior myocardial infarction and 21.87/1,000 subject-years in subjects with heart failure. Risk decreased by 20% from the initial to the final time period, with a greater decrease observed in those with (25%) compared with those without (12%) clinical heart disease. Thus, CA incidence varied considerably across clinical groups. The results provide insights regarding absolute and population-attributable risk in clinically defined subgroups, information that may aid strategies aimed at reducing mortality from CA.