In a retrospective study, 54 patients with treatment-resistant major depressive disorder (TRD) completed a free-viewing task in which they had to freely explore pairs of faces (an emotional face (happy or sad) opposite to a neutral face). Attentional bias to emotional faces was calculated for early and sustained attention. We observed a significant negative correlation between depression severity as measured by the 10-item Montgomery-Åsberg Depression Rating Scale (MADRS) and sustained attention to happy faces. In addition, we observed a positive correlation between depression severity and sustained attention to sad faces. No significant correlation between depression severity and early attention was found for either happy or sad faces. Although conclusions from the current study are limited by the lack of comparison with a control group, the eye-tracking free-viewing task appears to be a relevant, accessible and easy-to-use tool for measuring depression severity through emotional attentional biases in TRD.