BACKGROUND: Early improvement in positive emotions-more than decreases in negative emotions-was highly predictive of treatment response in an ecologically valid prospective manner. This result needs replication with simpler assessments to determine whether it can be translated into clinical practice. METHODS: 2049 adult depressed outpatients receiving agomelatine were assessed at inclusion, week 2, and week 6 using the clinician-rated Quick Inventory of Depressive Symptomatology, Sheehan Disability Scale, Clinical Global Impression scale, and Multidimensional Assessment of Thymic States (MATHYS), an auto-questionnaire rating the frequency of emotions, including sadness and joy, over the previous week. RESULTS: Joy and sadness had a relatively low correlation coefficient at baseline (r=-0.277), joy (r=-0.160) being less correlated with clinical severity than sadness (r=0.317). An increase in joy at week 2 had higher specificity (85.04%) and positive predictive value (70.55%) for treatment response than decreased sadness (57.92% and 66.04%, respectively), and the global capacity of the former to predict remission, either clinical (Yule Q coefficient, 39.96%) or functional (44.35%), was even better compared to the prediction of clinical response (37.38%). LIMITATIONS: MATHYS retrospectively assesses emotions, with five possible ratings only, relying on self-rated frequencies. With only a 6-week follow-up, conclusions are limited to short-term aspects of clinical and functional remission. CONCLUSIONS: Early improvement in joy during the first 2 weeks of treatment is strongly specific for treatment response and remission. The frequency of joy captures the predictivity and may deserve further study regarding inclusion in depressive rating scales.