Drugs have side effects that manifest as signs or symptoms which are sometimes undistinguishable from signs or symptoms of active disease. The conventional approximation of the rate of side effects of drugs is by subtracting the rate of signs and symptoms in the placebo group from that in the drug group. This measures net side effects and is adequate in studies with healthy volunteers, in which no interaction between drug and disease exists. For ethical and practical reasons, however, volunteer studies cannot be large and the frequency of non-rare side effects must be estimated in large-scale clinical trials. In the latter, biasing drug disease interactions may occur. We report on such a hitherto undescribed interaction: the pharmacological clinical activity bias. If one is interested in estimating not the net, but the direct or intrinsic, ie, drug-attributable side effects, the conventional approximation is biased whenever, in clinical trials, both of two conditions apply. The first is that the variable on the scale of which a sign or symptom is recorded as a putative side effect, is also in the absence of drug affected by uncontrolled disease. The second is that the drug has pharmacological clinical activity (A) on that sign or symptom, thus reducing the contribution of disease (D) to what is measured. In this case the drug affects the variable under study both directly, through its intrinsic side effect, and indirectly, through its clinical activity, and the rate of attributable side effects differs from the rate of net side effects as calculated by the conventional approximation. We present a simple deterministic model, which assumes that disease remains stable if untreated, additivity of the relative contributions of drug, placebo and disease to the total rate of the sign or symptom, and no other interaction between intrinsic properties of the drug and active disease than pharmacological clinical activity. This theoretical model quantifies the bias as DO(Ad-Ap), in which DO is the baseline frequency of the sign or symptom in the studied patients, and Ad and Ap are the intrinsic clinical activities of drug and placebo, respectively, on the sign or symptom under study. The model confirms that the conventional approximation of drug side effects is unbiased only in healthy volunteers or with drugs devoid of clinical activity. Without correction by such a model, any clinical activity of the drug or manifestation of active disease will cause the conventional approximation of side effects to be biased. This may manifest as artifacts such as attribution of a side effect when there is none, and as under- or overestimation, pseudotachyphylaxis, or pseudo-delayedness of attributable side effects.