Efficiency is an important indicator of an industry's development. To explore the temporal dynamics of tourism efficiency, this paper attempts to discuss its evolutional models, based on data envelopment analysis and the Malmquist Index (MI) of the seven-year panel data for 21 cities. The results show that, first, the province as a whole has relatively high tourism efficiency. Variations among cities, however, are quite large. Second, MI shows a trend of improvement in tourism efficiency, although there are differences among cities. Third, based on the magnitude and fluctuation of tourism efficiency, a quadrant chart is constructed on which cities can be classified into four categories. Fourth, based on the trajectories of corresponding points of each city on the quadrant chart, four tourism efficiency evolutional models can be identified: stable model, reciprocating model, progressive model and radical model. These findings provide a new perspective for tourism research and management practice.