In order to reveal the temporal changes, spatial differences and sources, dynamic evolution and convergence of tourism eco-efficiency in Xinjiang, China, the MinDS model of undesired output was used to measure its efficiency level, ArcGIS technology, Dagum Gini coefficient and its decomposition method and kernel density were used. Estimation and other tools were used to explore its spatial and temporal differentiation pattern from 2005 to 2019, and a convergence model was constructed to verify its spatial convergence. The results show that the ecological efficiency of Xinjiang’s tourism industry presents a temporal and spatial distribution evolution pattern of “high in the north and low in the south” that fluctuates steadily. There are β absolute convergence, β conditional convergence and club convergence, and the conditional convergence speed is significantly faster than absolute convergence speed. In the past 15 years, the proportion of cities in Xinjiang at level I has tended to increase, and the overall Gini coefficient has the characteristics of periodicity and volatility, and regional differences are the main source of spatial differences in tourism eco-efficiency. The tourism eco-efficiency between Xinjiang as a whole and northern Xinjiang has the characteristics of shrinking absolute differences and polarization, while the absolute differences in southern and eastern Xinjiang are expanding. Therefore, coordinating the eco-efficiency of tourism in the southern, northern and eastern Xinjiang regions in the future is the focus of work to achieve the high-quality development of Xinjiang’s overall tourism.