Depression is a debilitating condition with a high prevalence, but aetiology and pathophysiology are still unclear. Various reward-learning paradigms have been used to show impairments in depression. Both trait pessimism and neuroticism are associated with depression, but their link with the impairments in reward learning and decision-making have not been investigated. A Pavlovian conditioning task was performed by 32 subjects, 15 with depression. Participants had to estimate the probability of some fractal stimuli to be associated with a binary reward, based on a few observations. They then had to make a choice between one of the observed fractals and another target for which the reward probability was explicitly given. Computational modelling was used to succinctly describe participants' behaviour. Patients performed worse than controls at the task. Computational modelling revealed that this was caused by behavioural impairments during both learning and decision phases. Neuroticism scores across participants were significantly correlated with participants' inability to follow their internal value estimations. Our results demonstrate behavioural differences in probabilistic reward learning between depressed patients and healthy controls. Neuroticism was associated with the impaired ability to follow internal reward values and consequently with worse decision-making.A common symptom during depressive episodes is "bleak and pessimistic views of the future" [41]. The theory of learned helplessness posits that people with a pessimistic explanatory style (attributing their helplessness to a stable, global, internal cause) are at greater risk of developing depression [1]. There exists extensive evidence that patients diagnosed with MDD exhibit features of Beck's Negative Cognitive Triad, which is characterized by negative and pessimistic views about oneself, the world and the future [6], consistent with a pervasive pessimistic cognitive bias. The Beck Depression Inventory (BDI [5]) and the Beck Hopelessness Scale (BHS [4]) both measure aspects of this triad and Cognitive Behavioural Therapy (CBT), which targets these negative biases can be an effective treatment for depression [3,12]. Currently, however, the supporting evidence is based almost entirely on subjective clinical interviews and rating scales with little objective behavioural evidence. Here we addressed this issue of subjectivity using a novel experimental paradigm and computational models of decision-making.We used a probabilistic reward-learning task, which has previously been reported to demonstrate individual behavioural differences that were associated with Life Orientation Test -Revised (LOT-R; measuring optimism) scores [36], as well as neuroticism scores (see Methods and Materials and Supplement) in healthy people. In the task, participants were asked to maximize their rewards by choosing between fractal stimuli, for which they could estimate the probability of reward from previous passive observations, and another target associated with an explicit...