||Accelerating urbanization has created tremendous pressure on the global environment and energy supply, making accurate estimates of energy use of great importance. Most current models for estimating electric power consumption (EPC) from nighttime light (NTL) imagery are oversimplified, ignoring influential social and economic factors. Here we propose first classifying cities by economic focus and then separately estimating each category’s EPC using NTL data. We tested this approach using statistical employment data for 198 Chinese cities, 2015 NTL data from the Visible Infrared Imaging Radiometer Suite (VIIRS), and annual electricity consumption statistics. We used cluster analysis of employment by sector to divide the cities into three types (industrial, service, and technology and education), then established a linear regression model for each city's NTL and EPC. Compared with the estimation results before city classification (R2: 0.785), the R2 of the separately modeled service cities and technology and education cities increased to 0.866 and 0.830, respectively. However, the results for industrial cities were less consistent due to their more complex energy consumption structure. In general, using classification before modeling helps reflect factors affecting the relationship between EPC and NTL, making the estimation process more reasonable and improving the accuracy of the results.