To address the retrieval errors in passive satellite remote sensing of aerosol parameters due to the uncertainty of aerosol models, a novel aerosol layer height retrieval algorithm based on Bayesian theory is introduced and applied to the TROPOspheric Monitoring Instrument (TROPOMI) of the Sentinel-5 Precursor (Sentinel-5P) satellite in this work. The algorithm determines the aerosol model that meets the current observation data conditions based on the model evidence (conditional probability density of aerosol models) of different candidate aerosol models, and obtains the estimated maximum and estimated mean values as the results by two model selection schemes, respectively. Taking a real wildfire event observed by TROPOMI as an example, the retrieval results show a good spatial agreement with the official products. The underestimation found in previous algorithms is significantly improved, which proves that the algorithm can efficiently select a suitable aerosol model in the lack of a prior knowledge, and will offer a new solution for future operational data processing of aerosol layer height inversion from hyperspectral satellites.