Unipolar and bipolar depression are episodic, recurrent illnesses for the majority of patients. Because each episode engenders considerable costs for patients, families, and society, prevention of recurrences has high priority. Numerous studies demonstrate that maintenance antidepressants or mood stabilizing medications are efficacious in preventing recurrences. A review of maintenance studies supports the view that all antidepressants perform significantly better than placebo in preventing recurrences of depression--with the stipulation that full antidepressant doses be employed. Earliest studies, conducted two decades ago, evaluated tricyclics (TCAs), heterocyclics, and lithium, while recent studies have focused on selective serotonin reuptake inhibitors (SSRIs). Compliance is essential. Strategies for enhancing compliance include selection of medications with reported safety and few side effects, education of patients and families, referral to patient advocacy groups, and use of new technological compliance aids. Preliminary data suggest that SSRIs are better tolerated than TCAs; fewer patients discontinue these agents due to side effects. Selection criteria for maintenance treatment have not been well determined, but three or more prior episodes is recognized as a relatively strong indicator. Other clinical or genetic criteria have also been suggested. For various reasons, patients may discontinue medications, and when this happens withdrawal phenomena may occur. Withdrawal effects are well documented for all antidepressants and can be profound with TCAs. After stopping some SSRIs, a few withdrawal symptoms may have similarities with those following discontinuation of TCAs, but unique "CNS-like" effects are frequently described, including brief recurrent episodes of dizziness, lightheadedness, vertigo, electric shock-like sensations, and gait instability. These appear to be half-life dependent, with agents with shorter half-lives having more discontinuation symptoms. If antidepressant medications must be discontinued, a gradual taper is preferable, perhaps extending three to six months or longer to prevent discontinuation effects, enable adaptation at the receptor level and allow earlier recognition and treatment of recurrent depressive symptoms.