Current antidepressant agents are similar in efficacy to the original drugs discovered in the 1950s. The development of new treatments for depression is, however, limited by the absence of validated human biomarker models to predict efficacy, clinical profile and dosing. Such models need to meet key criteria for biomarkers including sensitivity, specificity and relevance to depression. Here we review studies exploring whether early changes in emotional processing with antidepressant drug administration meet these criteria. A large body of evidence suggests that changes in emotional memory are particularly relevant to depression and to antidepressant drug action whereas changes in attentional processing are sensitive to anxiolytic drugs. These tasks are not consistently affected by agents which have failed in clinical trials in depression, but do show changes in the predicted direction with agents associated either with amelioration or induction of symptoms. Hence, early assessment of novel drugs on emotional processing may predict likely clinical effects and dosing prior to randomized controlled trials. Greater validation is required to assess whether these effects are an obligatory component of effective treatment of depression and whether use of these models can improve the accuracy of go/no-go decisions in drug development.