Locust outbreaks are major biological disturbances in grassland ecosystems of arid and semi-arid regions. Accurate identification of locust suitable habitats is important for regional monitoring and management. However, direct comparisons between multi-criteria analysis (MCA) and species distribution models (SDMs) under a unified framework remain limited. In this study, we compared these two approaches for dominant locust species in Xinjiang, China, including Calliptamus italicus, Gomphocerus sibiricus, and Locusta migratoria manilensis. We used the same environmental variables and occurrence records for all models. The MCA methods included the analytic hierarchy process (AHP), technique for order preference by similarity to ideal solution (TOPSIS), and ordered weighted averaging (OWA). The SDMs included the generalized linear model (GLM), maximum entropy model (MaxEnt), extreme gradient boosting (XGBoost), and an ensemble model. The results showed that SDMs had higher area under the receiver operating characteristic curve (AUC) and true skill statistic (TSS) values than MCA under the internal point-based evaluation framework, although both approaches effectively identified locust-suitable habitats. The two approaches also showed high spatial agreement in moderately and highly suitable habitats, with Jaccard indices of 0.88-0.92, and consistently identified the northern slopes of the Tianshan Mountains, the Ili River Valley, and the margins of the Junggar Basin as core suitable areas. These results indicate that the two approaches are complementary for locust monitoring and management.