Prospective population studies of incident heart failure (HF) are often limited by difficulties in assembling HF-free cohorts. In this study, public-use copies of the Cardiovascular Health Study (CHS) data sets were used to determine the sensitivity, specificity, and positive and negative predictive values of orthopnea and paroxysmal nocturnal dyspnea (PND), with and without the use of medications used in CHS HF criteria (diuretics plus digoxin or vasodilators), in the diagnosis of prevalent HF and in the assembly of a relatively HF-free population. Of the 5,771 community-dwelling older adults aged > or =65 years, 803 had orthopnea, 660 had PND, 1,075 had either symptom, 388 had both symptoms, 547 were using HF medications, and 4,315 had neither symptom and were not using HF medications. Definite HF was centrally adjudicated in 272 participants. The sensitivity, specificity, and positive and negative predictive values for either orthopnea or PND were 52% (95% confidence interval [CI] 46% to 58%), 83% (95% CI 82% to 84%), 13% (95% CI 11% to 15%), and 97% (95% CI 97% to 98%), respectively, and those for either orthopnea or PND or the use of HF medications were 77% (95% CI 72% to 82%), 77% (95% CI 76% to 79%), 14% (95% CI 13% to 16%), and 99% (95% CI 98% to 99%), respectively. In conclusion, only <20% of those with either orthopnea or PND had definite HF, which limits their usefulness in the diagnosis of prevalent HF in the community. However, nearly 99% (negative predictive value) of those with neither symptom nor using HF medications also did not have HF, which may be useful as a simple and inexpensive tool in assembling relatively HF-free cohorts for prospective population studies of incident HF.