This paper deals with the application of an AI language, based on mechanical theorem prover (PROLOG), to deduce drug interactions. One hundred and thirty-one drug interactions were predicted by logical deduction with a program system comprising about 52 drugs in practice and 32 independent possibilities of deduction. The database of the system contains the chemical and pharmaceutical characteristics of the active components of the drug preparations, and the deductions are based on general pharmacokinetic, pharmacodynamic or other biochemical relationships concerning interactions.