The classification of forest ecosystems is fundamental for identifying forest ecological processes. Traditional methods of structural classification, such as climatic, geo-hydrologic, vegetative, and eco-systematic methods, reflect only external forest features without distinguishing essential functional differences. The functional classification of forest ecosystems would help remedy the deficiencies of traditional structural classification methods and provide a theoretical basis for forest management. Nutrient cycling, which is one of the main functions of forests, plays an important role in protecting the stability and sustainable development of forest ecosystems. An attempt to classify a forest ecosystem according to nutrient cycling processes is certainly a significant work. In this paper, nutrient cycling characteristics of 18 different forest ecosystems in 3 regions of the Loess Plateau were analyzed. The regions included the sand-blown region in the north part of the Loess Plateau, the hilly region in the middle part of the Loess Plateau, and the gullied region in the south part of the Loess Plateau. Indices of nutrient accumulation and distribution (including biomass, litter accumulation, and nutrient accumulation), nutrient cycling flux (annual absorption, retention, and return), and nutrient cycling efficiency (recycling coefficient, utilization ratio, and nutrient productivity) were calculated. Self-organizing Feature Maps (SOFM) was used for nutrient cycling classification. Through the analysis, two first-order classes (including I-Moderate-quick nutrient cycling and II-Slow nutrient cycling) and six second-order types were identified. The classification results were consistent with the actual forest characteristics. Furthermore, the results accurately reflected differences in nutrient cycling characteristics among the different ecosystems. A variety of indices reflecting different nutrient cycling processes were used in this classification. This approach avoided errors and made the results more reasonable. Finally, the feasibility of using SOFM for the classification of nutrient cycling in forest ecosystems has been demonstrated through this study.