<p>In existing pre-disaster earthquake reconstruction cost estimation methods, the basic assumption is that the building meets building code standards. However, in Indonesia, many buildings do not meet these standards, requiring a special approach. One such approach is to categorize buildings based on damage levels that follow standardized damage criteria. The main objective of this research is to identify the cost drivers of earthquake rehabilitation and reconstruction at each level of damage. This research uses multiple linear regression analysis models for each damage level (light, medium, and heavy). The regression analysis was conducted on 79 public buildings (schools, clinics, and government buildings) from post-earthquake reconstructions in Lombok in 2018 and Mamuju in 2021, Indonesia. The results show that, at the light damage level, variable cost drivers were identified as seismicity, building occupancy level, total floor area, reconstruction duration, and total reconstructed ceiling area. At the moderate damage level, the identified variable cost drivers were seismicity, building occupancy level, reconstruction duration, total reconstructed wall area, and demolition cost per total area. At the heavy damage level, the variable cost drivers identified were seismicity, location class, structure type, total floor area, and total reconstructed wall area. Identifying cost drivers is important for improving the accuracy of pre-disaster estimation models. In addition, the identified cost driver variables also reflect the key variables in the standardized building code that are often not complied with in Indonesia, indicating that regulatory improvements could begin with these variables.</p>