In this conceptual review we argue that by certifying some of the atypical neuroleptics (or, if one prefers, antipsychotics) as indicated for the 'maintenance' treatment of bipolar disorder, the US FDA has created confusion in the field. These maintenance indications are based on studies using a 'relapse prevention' design, a design that does not address whether the agents tested can prevent new episodes of illness, i.e. recurrence prevention or true prophylaxis. We found that the relapse prevention design fails to prove that these agents are mood stabilizers because patients are pre-selected to respond to the study drug for an acute mood episode (mania) and when they relapse, they do so into an episode of the same polarity (i.e. mania). We believe that this represents withdrawal into the same mood episode that patients experienced before the maintenance study began, rather than prevention of a new mood episode, as research into the natural history of bipolar disorder indicates that such new episodes typically are of the opposite polarity. Thus, the inability of neuroleptics to prevent depression in such maintenance studies reflects the general inability to prevent any new mood episode recurrence (which we believe should be defined as 6 months or longer after the index episode). If one defines a mood stabilizer, as we do, as a drug that prevents new episodes of mania and depression in monotherapy, then these studies do not show that atypical neuroleptics are mood stabilizers. Future maintenance research studies in bipolar disorder should use the prophylaxis design (i.e. without pre-selection of drug responders), rather than the relapse prevention design.