Depression widely affects global populations and is one of the leading causes of disability and suicide. Despite its prevalence, traditional diagnosis for depression is exceedingly associated with misidentification and over-estimation, due to its subjective nature. With advances in affective computing, computational approaches make it possible to discern depression through second party physiological indicators; people observing the behaviour of depressed individuals have measurable changes in their physiological signals.We explored Blood volume pulse (BVP), Galvanic Skin Response (GSR), Skin Temperature (ST) and Pupillary Dilation (PD) from observers as valid sources to indicate depression in others. The behaviour of individuals suffering from four levels of depression was shown in 16 videos to 12 experimental observers whose physiological signals were recorded. We found that depression provokes visceral physiological reactions in observers that we can measure, resulting in neural network classification of 94% accuracy. In contrast, we also found that depression does not provoke strong conscious recognition ('verbal') in observers, which is only slightly over a chance level, at 27%.