Abstract: Taking 28 state-level poverty-stricken counties in Guangxi as examples, three-stage DEA model is used to measure and evaluate the poverty alleviation efficiency of tourism in 28 poverty-stricken counties by selecting the time cross-section data of 2011, 2015 and 2018. The results show that: (1) Before and after removing the interference factors, there are significant differences in the comprehensive efficiency of poverty alleviation through tourism and its decomposition efficiency, which indicates that the measurement of poverty alleviation efficiency through tourism by using the traditional DEA model will lead to estimation errors. (2) After the adjustment, the comprehensive efficiency of tourism poverty alleviation in 28 counties in Guangxi showed an upward trend in 2011, 2015 and 2018, with an average of 0.651, 0.777 and 0.853, respectively. The differences of pure technical efficiency among poor counties are great, and the internal heterogeneity is prominent. The overall scale efficiency is fluctuating and rising, which is the main driving force for the improvement of tourism poverty alleviation efficiency in ethnic minority areas. (3) The spatial distribution of poverty alleviation efficiency of poverty-stricken counties in 28 countries changed from "high dispersion and low concentration" to "high in the east and west and low in the northwest". The evolution of scale efficiency and comprehensive efficiency of poverty alleviation by tourism was similar, while the spatial evolution of pure technical efficiency was different from both. (4) The number of counties with "double-high" tourism poverty alleviation efficiency is the largest, followed by "double-low" tourism poverty alleviation efficiency, which indicates that the overall effect of poverty alleviation assisted by tourism in poverty-stricken counties in Guangxi is remarkable, but some regions still need to increase the input of tourism industry elements and improve the production technology and management level.