The objective of this study was to evaluate, for patients at risk of a new depressive episode, the net cost of maintenance therapy with milnacipran compared with a symptomatic treatment of further episodes. Using clinical decision analysis techniques, a Markov-state transition was constructed to estimate the 12 months direct costs of the two therapeutic strategies. Model construction and probabilities for performing the analysis were primarily based on a controlled phase III clinical trial demonstrating the prophylactic efficacy of milnacipran compared to placebo. Others parameters and unit costs were obtained from published sources. For each group (maintenance and episodic treatment groups), the model simulated the clinical evolution of patients on 6 successive 2-months cycles. Costs were affected for each health state period (remission, depression, abandonment of health care, suicide). The baseline analysis showed the mean costs per patient and year were 627 FF (105 US$) higher for maintenance treatment. Sensitivity analysis suggested that costs were equal under a 25% rate of hospitalization hypothesis for a depressive episode. Maintenance costs were 1,587 FF (265 US$) lower than episodic treatment costs for depressed subjects with a good initial response to milnacipran (HDRS-21 score at remission < 5); this economic benefit remained under a lower rate of hospitalization hypothesis (12%). Based on the study assumptions, maintenance treatment with milnacipran appears to be clinically and economically justified for patients at high risk of hospitalization when having a recurrence, and even more for patients with an excellent initial acute response.