Medication nonadherence is a major obstacle to translating treatment efficacy from research settings into effectiveness in clinical practice for patients with affective disorders. Adherence to beneficial drug therapy is associated with lower mortality compared with poor adherence. Reduced adherence is associated with increased suicide risk, especially when lithium is discontinued. The aim of this paper is to review the prevalence, predictors and methods for improving medication adherence in unipolar and bipolar affective disorders. Studies were identified through Medline and PsycInfo searches of English language publications between 1976 and 2009. This was supplemented by a hand search and the inclusion of selected descriptive articles on good clinical practice. Estimates of medication nonadherence for unipolar and bipolar disorders range from 10 to 60% (median: 40%). This prevalence has not changed significantly with the introduction of new medications. There is evidence that attitudes and beliefs are at least as important as side effects in predicting adherence. The limited number of empirical studies on reducing nonadherence indicate that, if recognized, the problem may be overcome. Clinical data highlight the importance of extended courses of medication in improving the long-term prognosis of patients with affective disorders.