Deep analysis of the characteristics of bus load is of great significance for bus load forecasting accuracy inprovement, power network safety and stability evaluation, and demand response potential identification. A data-driven based analytical framework of bus load characteristic is proposed in this paper, which is different from traditional load ratio and peak time. On the basis of data cleaning and normalization, the analysis of the daily bus load curve is carried out by clustering algorithm based on the Mahalanobis distance. Then, based on the clustering results, four indexes, including pattern switch entropy, relative volatility, daily average load, and temperature sensitivity are put forward to describe bus load characteristics. Simulation results of 130 bus load data in Guangzhou show that the proposed indexes can better describe the bus load characteristics, and can achieve better classification results.