The restoration of degraded forest has been a hot topic in forestry and ecology. As a basic structural unit of soil, soil aggregate composition and stability are important indicators of soil fertility and quality, and the formation of soil aggregate relies on aggregate bonding materials. However, the relationship between aggregate bonding materials and aggregate stability is uncertain. To explore the effects of different restoration approaches of degraded subtropical forests on aggregate stability and its underlying mechanisms, we compared the secondary forest (natural restoration) as a control (CK), with three different common subtropical plantations which were: (1) first-generation Cunninghamia lanceolata (P1), (2) second-generation C. lanceolata (P2), and a Pinus. taiwanensis plantation (P3). We measured the aggregate stability and the contents of different aggregate bonding materials, and then analyzed the influence of these materials on soil aggregate stability. Results showed that forest restoration approach significantly affected soil pH, C/N, available phosphorus, as well as the composition and stability of aggregate. Among all the forests, the aggregate stability was significantly lower in CK than in the other three plantations, and P2 had the greatest proportion of large aggregates and the highest aggregate stability. The forest restoration approach also significantly affected soil free Fe oxide content, mycorrhizal density, and easily extractable glomalin-related soil protein (EEG). Aggregate stability was positively correlated with free Fe oxide content and mycorrhizal density, but negatively correlated with EEG. For subtropical degraded forests, the results indicated that artificial restoration was better than natural restoration to enhance soil aggregate stability, with soil free Fe oxide, mycorrhizal density and EEG being the main bonding materials that significantly affected the aggregate stability.