A sequential approach is presented to the problem of determining the minimum number of blood samples needed to calculate the plasma to saliva concentration ratio to a required precision. The method was applied to salicylate concentrations obtained from six rheumatoid arthritis patients. In order to achieve a 10 per cent coefficient of variation in the plasma to saliva salicylic acid concentration ratio, on average 9 samples were required for total plasma concentration and 8 samples for unbound concentration. In some cases it was not possible to achieve the required precision with the given number of samples. Correlation of salicylic acid concentrations in saliva with total and unbound plasma concentration were equally as good. The limitations of saliva data in clinical pharmacokinetic studies are discussed.