Many studies have found correlations between reflective thinking and disbelief in God or lower levels of religiosity — so-called analytic atheism. Some data even detect analytic atheism in highly reflective people like philosophers. However, there is growing evidence that the most widely used tests of reflection confound reflection with ancillary abilities such as numeracy. Moreover, some studies do not detect analytic atheism in every country. Further, experimentally encouraging reflection has made some non-believers more open to believing in God. Finally, there is growing evidence of low data quality from one of the most common sources of online research participants. These prior results indicate that analytic atheism may be less than universal or partially explained by confounding factors. To address these possibilities, we developed better measures of religiosity and reflection, controlled for more confounds, and studied both paid and ad-recruited participants (N > 70,000). All four studies detected signs of analytic atheism above and beyond confounds. We also discovered analytic apostasy: the better a person performed on reflection tests, the greater their odds of losing their religion since childhood — even when controlling for confounds. Religious conversion was rare, but the final study recruited enough converts to find that the odds of becoming religious since childhood were probably unrelated to reflection, suggesting reflection’s relationships to conversion and deconversion are asymmetric. Detected relationships were usually small, but few failed to reach conventional levels of significance, indicating that reflective thinking may be a reliable albeit marginal predictor of apostasy.