The Earth occultation technique has broad applications in both astronomy and atmospheric density measurements. We construct the background model during the occultation of the Crab Nebula observed by the Insight-Hard X-ray Modulation Telescope (Insight-HXMT) at energies between 6 and 100 keV. We propose a Bayesian atmospheric density retrieval method based on the Earth occultation technique, combining Poisson and Gaussian statistics. By modeling the atmospheric attenuation of X-ray photons during the occultation, we simultaneously retrieved the neutral densities of the atmosphere at different altitude ranges. Our method considers the correlation of densities between neighboring atmospheric layers and reduces the potential systematic bias to which previous work may be subject. Previous analyses based on light-curve fitting or spectral fitting also lost some spectral or temporal information of the data. In contrast to previous work, the occultation data observed by the three telescopes on board Insight-HXMT is fully used in our analysis, further reducing the statistical error in density retrieval. We apply our method to cross-check the (semi)empirical atmospheric models, using 115 sets of occultation data of the Crab Nebula observed by Insight-HXMT. We find that the retrieved neutral density is ∼10%, ∼20%, and ∼25% less than the values of the widely used atmospheric model NRLMSISE-00, in the altitude range of 55–80 km, 80–90 km, and 90–100 km, respectively. We also show that the newly released atmospheric model NRLMSIS 2.0 is generally consistent with our density measurements.