We studied within-person variation over time in serum concentrations of five carotenoids. In a diurnal study involving 33 subjects, only the 1700 h blood samples demonstrated carotenoid concentrations different from the original 0800 values. Correlations between serum concentrations of the same carotenoids drawn 1 d apart ranged from 0.93 to 0.98. In a seasonal study involving 29 subjects, no systematic trends were observed for serum concentrations of these carotenoids. Correlations between concentrations of the same carotenoids drawn 1 y apart ranged from 0.57 to 0.82. Concentrations of different carotenoids within an individual tended to be correlated with each other. Obtaining one blood sample from subjects is a relatively imprecise way to estimate their usual serum concentrations of carotenoids. If an epidemiological study was to be based on only one determination of serum carotenoids, within-person variability in serum concentrations would attenuate true regression coefficients by 4-13% and would increase the required numbers of study subjects by 19-65%.