Accident analysis and prevention are helpful to ensure the sustainable development of transportation. The aim of this research was to investigate the factors associated with the severity of low-visibility-related rural single-vehicle crashes. Firstly, a latent class clustering model was implemented to partition the whole-dataset into a relatively homogeneous sub-dataset. Then, a spatial random parameters logit model was established for each dataset to capture unobserved heterogeneity and spatial correlation. Analysis was conducted based on the crash data (2014–2019) from 110 two-lane road segments. The results show that the proposed method is a superior crash severity modeling approach to accommodate the unobserved heterogeneity and spatial correlation. Three variables—seatbelt not used, motorcycle, and collision with fixed object—have a stable positive correlation with crash severity. Motorcycle leads to a 12.8%, 23.8%, and 12.6% increase in the risk of serious crashes in the whole-dataset, cluster 3, and cluster 4, respectively. In the whole-dataset, cluster 2, and cluster 3, the risk of serious crashes caused by seatbelt not used increased by 5.5%, 0.1%, and 30.6%, respectively, and caused by collision with fixed object increased by 33.2%, 1.2%, and 13.2%, respectively. The results can provide valuable information for engineers and policy makers to develop targeted measures.