Regional innovation is an important research topic in economic geography, the spatio-temporal evolution and mechanism of regional innovation efficiency have recently become a hot for economic geographers. From the perspective of input and output efficiency, this paper constructs evaluation indicator of regional innovation, with the help of Constant Returns to Scale (CRS) and Variable Returns to Scale (VRS) models, and Malmquist indicator method of Data Envelopment Analysis (DEA), to analyze regional innovation performance, evolution trend, spatial differentiation, and evolution mechanism of Yangtze River Delta Urban Agglomeration (YRDUA) of China. The results show that: (i) Innovation efficiency of YRDUA is generally low, most of which is less than 80 percent of optimal efficiency; however, it kept rising from 2000 to 2015. (ii) Spatial inequality of regional innovation in YRDUA is significant, with a spatial pattern in the shape of "Z", composed by Hefei, Nanjing, Shanghai, Hangzhou and Ningbo, innovation efficiency of Shanghai is higher than Zhejiang, Anhui and Jiangsu. (iii) Technology progress is the most important influencing factor, all kinds of changing indicator show a trend of rise, and the total factor productivity is changing significantly. This research can provide theoretical reference for the YRDUA to achieve high-quality integration.