Résumé : Objective: The timing and rate of improvement after the initiation of an antidepressant has implications for establishing the mechanism of antidepressant action and for answering the clinically relevant question of how long an appropriate trial of antidepressant medication should be. We explore the individual trajectories of relative change in depression severity to establish what proportion of individuals experience early and late onset of improvement. Method: Longitudinal latent class analysis was applied in a secondary analysis of data obtained from the Genome-Based Therapeutic Drugs for Depression (GENDEP) study. In the GENDEP trial, conducted in 9 European academic psychiatry centers from July 2004 to June 2008, 811 treatment-seeking adult subjects with DSM-IV major depression received escitalopram or nortriptyline for 12 weeks. Montgomery- Asberg Depression Rating Scale measurements were taken weekly. The secondary analysis reported in this article was conducted in 2010. Results: A model with 9 latent classes provided a good description of the individual trajectories of symptom change over time. These classes included 3 nonresponder classes, 3 classes with varying degrees of improvement concentrated in the first 3 weeks (early improvement), and 3 classes with varying degrees of improvement that was more prominent in the second 3 weeks than in the first 3 weeks (delayed improvement). More than half of the subjects who eventually reached remission showed a pattern of delayed improvement, and their eventual outcome could not be predicted from early time points. Early marked response occurred more frequently in subjects treated with nortriptyline than in those treated with escitalopram (12.9% vs 7.5%, X 2 = 6.29, P = .01). Delayed complete remission occurred more frequently in subjects treated with escitalopram than in those treated with nortriptyline (13.6% vs 6.1%, X 2 = 11.52, P = .0007). Conclusions: Both early and delayed improvement are common. Although early changes are maintained, the eventual outcome of 12-week antidepressant treatment can be accurately predicted only after 8 weeks. Trial Registration: http://www.controlled-trials.com Identifier: ISRCTN03693000. © Copyright 2011 Physicians Postgraduate Press, Inc.