As clinical experience with lithium treatment of bipolar disorders accumulates, factors predictive of nonresponse are emerging. Prominent among these are conditions such as comorbid substance abuse and the presence of the malignant variants rapid cycling and mixed states. Lithium therapy is further complicated by noncompliance, attributable in large measure to burdensome side effects such as memory impairment and cognitive slowing. The issues of lithium nonresponse and noncompliance have driven the search for alternative agents, such as the anticonvulsants carbamazepine and valproic acid. While preliminary evidence suggests that these agents may provide improvements over lithium in terms of tolerability and response (especially in rapid cycling and mixed states), methodologic limitations temper the conclusiveness of these findings. The natural history of bipolar disorders is defined by characteristic symptom clusters which evolve in intensity and duration over long time scales, resulting in a high morbidity and mortality. These features raise critical concerns for research design, especially concerning the ethics of placebo controls and the need to generate long-term observational data upon which to predicate meaningful treatment recommendations. With the apparent increase in lithium nonresponse in recent years, the need to identify and definitively evaluate alternative agents is becoming imperative.