A mechanism of anesthesia is proposed that addresses one of the most troubling peculiarities of general anesthesia: the remarkably small variability of sensitivity within the human population and across a broad range of animal phyla. It is hypothesized that in addition to the rapid, saturable binding of a neurotransmitter to its receptor that results in activation, the neurotransmitter also acts indirectly on the receptor by diffusing into the postsynaptic membrane and changing its physical properties, causing a shift in receptor conformational equilibrium (desensitization). Unlike binding, this slower indirect mechanism is nonspecific: each neurotransmitter will, in principle, affect all receptors in the membrane. For proteins modeled as having only resting and active conformational states, time-dependent ion currents are predicted that exhibit many characteristics of desensitization for both inhibitory and excitatory channels. If receptors have been engineered to regulate the time course of ion currents by this mechanism, then (a) mutations that significantly alter receptor sensitivity to this effect would be lethal and (b) by design, excitatory receptors would be inhibited, but inhibitory receptors activated, so that their effects are not counterproductive. The wide range of exogenous molecules that affect the physical properties of membranes as do neurotransmitters, but that do not bind to receptors, would thus inhibit excitatory channels and activate inhibitory channels, i.e., they would act as anesthesics. The endogenous anesthetics would thus be the neurotransmitters, the survival advantage conferred by their proper membrane-mediated desensitization of receptors explaining the selection pressure for anesthesic sensitivity.