The current meta-analysis examines the cognitive approach to lie detection. Our goal was to assess the practical utility of this approach by examining whether it improves the lie detection ability of human observers. The cognitive approach to lie detection led to an average accuracy rate of 60.00%, 95% CI [56.42; 63.53] and a bias corrected average accuracy rate of 55.03%, 95% CI [48.83; 61.16]. Critically, this result is moderated by whether observers were informed, or not, about which cues to focus on. Naïve observers had average accuracy rates of 52.37%, 95% CI [48.80%; 55.93%], little better than chance. In contrast, informed observers had average accuracy rates of 75.81%, 95% CI [71.52%; 79.86%]. This promising result is qualified by indications of publication bias, considerable heterogeneity between studies, and a lack of research on important practical issues, such as the influence of counter-measures. Although these shortcomings raise a note of caution, we remain optimistic about future research on the topic.