Abstract Biofeedback is an acquisition technique of self-regulation ability of an autonomic function, of which we are normally unaware, through a series of training aided by an additional outer feedback pathway. We proposed a mathematical model of biofeedback in which a learning system on the conscious level learns characteristics of a subconscious regulation system corresponding to the biological function. When the learning converges, the learning system itself becomes an inverse system of the regulation system. Then, if a regulation command is put to the learning system on the conscious level, it drives the regulation system strictly following the command without the outer feedback pathway, which enables voluntary control of the biological function.