An urban spatial cluster (USC) describes one or more geographic agglomerations and the linkages among cities. USCs are conventionally delineated based on predefined administrative boundaries of cities, without considering the dynamic and evolving nature of the spatial extent of USCs. This study uses Defense Meteorological Satellite Program/Operational Linescan System (DMSP/OLS) nighttime light (NTL) satellite images to quantitatively detect and characterize the evolution of USCs. We propose a dynamic minimum spanning tree (DMST) and a subgraph partitioning method to identify the evolving USCs over time, which considers both the spatial proximity of urban built-up areas and their affiliations with USCs at the previous snapshot. China is selected as a case study for its rapid urbanization process and the cluster-based economic development strategy. Four DMSTs are generated for China using the urban built-up areas extracted from DMSP/OLS NTL satellite images collected in 2000, 2004, 2008, and 2012. Each DMST is partitioned into various subtrees and the urban built-up areas connected by the same subtree are identified as a potential USC. By inspecting the evolution of USCs over time, three different types of USCs are obtained, including newly emerging, single-core, and multicore clusters. Using the rank-size distribution, we find that large-sized USCs have greater development than medium- and small-sized USCs. A clear directionality and heterogeneity are observed in the expansions of the ten largest USCs. Our study provides further insight for the understanding of urban system and its spatial structures, and assists policymakers in their planning practices at national and regional scales.