Purpose This article provides a re‐analysis of Vrij et al.'s (2017, Leg. Crim. Psychol. 22, 1) meta‐analysis of the cognitive approach to lie detection. Vrij et al.'s analyses confounded dependent variables, capitalized on aberrant controls, and used unreliable data to inflate support. Methods Meta‐analysis was used to reanalyse Vrij et al.'s data. Studies of human detection and studies involving statistical classification were analysed separately. Results The advantage offered by the cognitive approach was much smaller than previously claimed. Accuracies in control conditions were unusually low, and the most supportive findings came from the least reliable data. Conclusions Human detection and statistical classification are different. The evidence for the cognitive approach has been overstated.