In order to test the predictions derived from emotional processing theory (EPT), this meta‐analysis examined correlations between outcome of exposure therapy and three process variables: initial fear activation (IFA), within‐session habituation (WSH) and between‐session habituation (BSH). Literature search comprised a keyword‐based search in databases, a reverse search and the examination of reference lists. Of the 21 studies included in the analyses, 17 provided data concerning IFA (57 endpoints, total N = 490), five concerning WSH (7 endpoints, total N = 116) and eight concerning BSH (22 endpoints, total N = 304). Owing to this data structure, analyses were performed using robust variance estimation with random‐effects models being assumed a priori. Results indicated that WSH and BSH are positively related to treatment outcome. By contrast, the statistical association between IFA and outcome of exposure was not confirmed, whereas our moderator analysis suggested that physiological process measures lead to higher correlations than non‐physiological ones. The results for IFA and BSH were affected by selective reporting. In sum, our results do not specifically strengthen EPT while matching other theoretical perspectives such as inhibitory learning and reality testing. Further research is needed to provide recommendations concerning the best way of delivering exposure therapy. Copyright © 2016 John Wiley & Sons, Ltd. Key Practitioner Message This meta‐analysis examined three variables of emotional processing theory (EPT). Initial fear activation was not linearly related to outcomes of exposure therapy. Habituation within and between sessions were shown to correlate with outcome. Outcome reporting bias was shown to play a crucial role in this meta‐analysis. Results do not specifically support EPT.