Transcranial magnetic stimulation (TMS) is a novel approved therapy for treatment-resistant depression. Little is known about its neurocognitive mechanisms of action. The existing literature has focused on resting-state neuroimaging. It therefore remains unknown what information processing changes TMS induces during treatment that drive mood change. Here we tested the hypothesis that transcranial magnetic stimulation (TMS) treatment changes emotional bias, increasing the focus on positive (versus negative) information processing. 49 patients with major depression received 20 daily sessions of open-label intermittent theta-burst TMS to left dorsolateral prefrontal cortex. Emotional bias was measured using behavioural and functional magnetic resonance imaging tasks of emotional face processing, both at baseline and after eight days of treatment. We tested whether early changes in these measures after the first week predicted clinical outcome at the end of treatment (4 weeks). As predicted, an increase in behavioural and neural measures of positive bias after one week predicted clinical response after four weeks of treatment. Behaviourally, response to TMS treatment was associated with a bias towards interpreting ambiguous facial expressions as positive. Neurally, clinical improvement was related to increased neuroimaging response for the contrast of positive versus negative emotional faces in rostral anterior cingulate cortex (rACC), and a positive bias in task-related functional connectivity between rACC and posterior default mode network. These early neurocognitive changes predicted clinical outcomes after four weeks of treatment, beyond early symptom reduction. Thus, clinical response to TMS treatment was linked to increases in positive bias in emotional processing early during treatment which might represent a neurocognitive mechanism of TMS depression treatment, potentially neurally distinct from antidepressant drugs.