Risk prediction functions for incident coronary heart disease (CHD) were estimated using data from the Atherosclerosis Risk in Communities (ARIC) Study, a prospective study of CHD in 15,792 persons recruited in 1987-1989 from four U.S. communities, with follow-up through 1998. Predictivity of which individuals had incident CHD was assessed by increase in area under ROC curves resulting from adding nontraditional risk factors and markers of subclinical disease to a basic model containing only traditional risk factors. We also assessed the increase in population attributable risk. The additional factors were body mass index; waist-hip ratio; sport activity index; forced expiratory volume; plasma fibrinogen, factor VIII, von Willebrand factor, and Lp(a); heart rate; Keys score; pack-years smoking; and subclinical disease marker carotid intima-media thickness. These factors substantially improved prediction of future CHD for men, less for women, and also increased attributable risks.