Abstract:City clusters are and high ground for national construction and regional economic development, and comprehensively exploring the differences in innovation efficiency of city clusters is of great theoretical and practical significance for narrowing the gap of innovation capacity between regions. To study the innovation efficiency of national-level city clusters from a dynamic perspective, we adopt the Pareto-optimal cross-efficiency model to comparatively analyze the urban innovation efficiency and its spatial dependence of the Pearl River Delta (PRD) city cluster and the Chengdu-Chongqing city cluster from 2010 to 2020, and analyze the factors influencing the innovation efficiency of city clusters by using the Tobit model. The results show that: the value of innovation efficiency of the PRD city cluster shows a downward trend, and the value of innovation efficiency of the Chengdu-Chongqing city cluster shows an upward trend, but in general, the innovation capacity of the PRD city cluster is still better than that of the Chengdu-Chongqing city cluster; at the spatial level, due to the implementation of the integration policy and the phenomenon of the siphoning effect, the Moran's I Index both show a positive and then a negative situation, which indicates that the influence of spatial location has been Moran's I index is positive and then negative, indicating that the influence of spatial location is gradually weakening. In terms of influencing factors, foreign-invested enterprises, population density, education expenditure have significant positive influence, the number of large-scale industrial enterprises and urban workers' pension insurance have significant negative influence, and ecological and environmental factors have a negative influence on the Pearl River Delta, but have no significant influence on the Chengdu-Chongqing urban agglomeration. To summarize, the urban innovation capacity of city clusters can be enhanced by focusing on creating highlands, filling in depressions and formulating locally adapted policies.