Both plasma and saliva cotinine levels have been reported in surveys of smoking behavior, and it is of interest to know how closely these two measures correspond. Plasma and saliva specimens were gathered from a sample of 605 respondents in the 1998 Health Survey for England and assayed for cotinine by a well-proven gas chromatographic method. Plasma and saliva cotinine concentrations were highly correlated (r=.99). On average, concentrations in saliva were 25% higher than in plasma, and this ratio applied both at the low levels attributable to passive smoking and across the range of active smoking values. The ratio was somewhat lower in younger people than in older people and also varied significantly by body mass index but did not differ by gender. Calculation of the limits of agreement revealed substantial uncertainty in the predicted plasma value corresponding to a given saliva cotinine, and vice versa. For comparisons across subjects, the mean plasma cotinine level corresponding to a mean saliva cotinine level can be estimated with confidence, but at the level of the individual, considerable predictive uncertainty remains.