As an important part of the smart expressway, the electronic toll collection (ETC) system generates massive real-time traffic data every day. In order to improve the quality of ETC data and classify user features, a highway user classification method based on ETC data was proposed, which included four parts: data optimization, trajectory association, quality evaluation and user feature classification. On the basis of analyzing and summarizing the ETC gantry and its data characteristics, a data extraction and optimization method was discussed. Based on the preprocessed toll data of the expressway road network, according to the differences of the travel characteristics of different users, a portrait labeling system for the travel characteristics of expressway users was developed from the aspects of vehicle operation status and transportation risk. Finally, combined with the relevant indexes of each type of vehicle, a highway user classification model was constructed, and the K-medoids algorithm was used to classify the driving vehicles into five categories. Based on the driving toll data of a province in May 2023, the proposed model is used to divide the vehicles into different categories, including medium-risk users of high-frequency and long-distance night travel, high-risk users of long-distance and high-risk travel on low-frequency weekends, medium-risk users of medium-frequency weekly morning peak travel, low-risk users of medium-frequency short-distance evening travel, and low-risk users of high-frequency location stability. This paper summarizes and analyzes the current status of the differentiated toll policy and implementation of expressways, and proposes a differentiated toll strategy based on user travel characteristics. At the same time, the ETC customer segmentation method provides innovative ideas for expressway operation and management units to explore precision marketing and hierarchical rate concessions for ETC customers.