A pronounced variability limits the usefulness of CYP1A2 phenotyping for drug therapy, for evaluating liver function, and for assessing the role of this enzyme in carcinogenesis. To identify and quantify sources of this variation, we estimated CYP1A2 activity in 863 healthy Caucasians using caffeine clearance derived from saliva concentrations before and 5–7 h after a caffeine test dose. Data from 786 individuals were eligible for evaluation (mean age 39 years, 415 women including 94 taking oral contraceptives, 401 non-smokers). Overall geometric mean (geometric SD) caffeine clearance was 1.34 ml min-1 kg b.w.-1 (1.65). The effect of the following covariates was evaluated by analysis of covariance: age, sex, oral contraceptives, body height, body weight, body mass index, number of cigarettes smoked, tar exposure from smoking, several indices of dietary caffeine consumption, intake of sauerkraut, and country of residence (Germany, Bulgaria or Slovakia). Estimated changes relative to arbitrarily defined basal caffeine clearance (male, nonsmoking, German resident) exerted by significant (P < 0.05) covariates were: coffee, 1.45-fold per litre of coffee drunk daily; body mass index, 0.99-fold per kg m-2; smoking, 1.22-fold, 1.47-fold, 1.66-fold, and 1.72-fold for 1–5, 6–10, 11–20, and > 20 cigarettes smoked per day, respectively; oral contraceptives, 0.72-fold; country of residence, 0.81-fold and 0.74-fold for Bulgaria and Slovakia, respectively; female, 0.90-fold. These covariates explained 37% of overall variation. The 95% confidence interval of individual clearance was 0.46–2.20 times the predicted value. No relevant polymorphism was found for CYP1A2 activity when adjusted for covariate effects.