Understanding energy users’ consumption patterns benefits both utility companies and consumers as it can support improving energy management and usage strategies. The rapid deployment of smart metering facilities has enabled the analysis of consumption patterns based on high-precision real usage data. This paper investigates data-driven unsupervised learning techniques to partition district heating users into separate clusters such that users in the same cluster possess similar consumption pattern. Taking into account the characteristics of heat usage, three new approaches of extracting pattern features from consumption data are proposed. Clustering algorithms with these features are executed on a real-world district heating consumption dataset. The results can reveal typical daily consumption patterns when the consumption linearly related to ambient temperature is removed. Users with heat usages that are highly imbalanced within a certain period of time or are highly consistent with the utility heat production load can also be grouped together. Our methods can facilitate gaining better knowledge regarding the behaviors of district heating users and hence can potentially be used to formulate new pricing and energy reduction solutions.