Studies of the mechanism of death in heart failure are dependent on the reliability and validity of classification of deaths as pump failure or arrhythmias (sudden). Two recent trials differed in that the Vasodilator Heart Failure Trial II (V-HeFT II) reported a higher incidence of sudden death than the Studies of Left Ventricular Dysfunction Treatment Trial (SOLVD) and an effect of enalapril on sudden death was not observed in SOLVD. A similar classification system was used in the two studies, but deaths in V-HeFT were classified centrally from a narrative summary, whereas deaths in SOLVD were classified in the field by individual investigators. To examine reliability, 10 narratives used to classify V-HeFT deaths were independently classified by 21 SOLVD investigators. In only 5 of 10 cases did 75% of SOLVD investigators agree with the V-HeFT classification. In no deaths were all SOLVD investigators in agreement on classification. Although V-HeFT classified 5 of 10 cases as sudden death, 16 of 21 SOLVD investigators classified less than 5 deaths as sudden and 1 classified none as sudden. The kappa statistic for interobserver agreement of 0.22 (P < or = .01) indicated interobserver agreement only slightly better than chance agreement. Therefore, the incidence of sudden death in heart failure is critically dependent on the bias of the investigator. Central classification will minimize inconsistencies, but it does not solve the problem that the mechanism of death is difficult to assign. Total mortality may be the only reliable endpoint in heart failure trials.