In the context of Lewy's phase delay hypothesis, the present study tested whether effective treatment of winter Seasonal Affective Disorder (SAD) is mediated by advancing of circadian phase. Following a baseline week, 78 outpatients with SAD were randomized into 8 weeks of treatment with either fluoxetine and placebo light treatment or light treatment and placebo pill. Depression levels were measured on the Ham17+7 and the BDI-II, and circadian phase was estimated on the basis of daily sleep logs and self-reported morningness-eveningness. Among the 61 outpatients with complete data, both treatments were associated with significant antidepressant effect and phase advance. However, pre- and post-treatment comparisons found that the degree of symptom change did not correlate with the degree of phase change associated with treatment. The study therefore provides no evidence that circadian phase advance mediates the therapeutic mechanism in patients with SAD. Findings are discussed in terms of the limitations of the circadian measures employed.