Electrocardiographic repolarization abnormalities characterized by T-wave morphology parameters such as the principal component analysis ratio and the relative and the absolute T-wave residuum (TWR(rel) and TWR(abs)) are predictive of cardiovascular and/or all-cause mortality. However, when using a "10-second median beat" for analysis, the reported mean values for TWR(rel) vary widely and parameter reproducibility is somewhat suspect. In repeated electrocardiographic recordings conducted 1 month and 1 year apart on 15 and 27 healthy individuals, respectively, we studied the said T-wave morphology parameters in single complexes and in reduced noise signal averages containing 10 and 200 complexes. Considering all subjects, the mean (+/-SD) TWR(rel) was highest in a single complex (0.0345% +/- 0.0183%), intermediate in the 10-beat signal-averaged complexes (0.0125% +/- 0.0051%), and lowest in the 200-beat signal-averaged complexes (0.0078% +/- 0.0036%) (P < .0001), with the same trend also observed in the TWR(abs) but not in the principal component analysis ratio. Reproducibility as quantified by within-subject variance and reliability as quantified by the intraclass (intrasubject) correlation coefficient also improved as the number of T-wave complexes analyzed increased. We conclude that signal averages consisting of more than 10 complexes (or more than 10 seconds worth of complexes) are required to produce reproducible and reliable values for TWR(rel) and TWR(abs).