Abstract: Innovation efficiency is the core element of urban innovation competitiveness. This paper constructs the assessment system of innovation efficiency based on input-output perspective and measures the innovation efficiency by DEA and Malmquist index models. Then the spatial and temporal evolution of innovation efficiency of the urban agglomerations in the middle reaches of the Yangtze River from 2006 to 2016 is analyzed. The GMM estimation is used to discover the spatial heterogeneity of the driving factors. The results show that: (1) In terms of the comprehensive technology efficiency of innovation, on one hand the innovation efficiency temporally increased from 0.731 in 2006 to 0.807 in 2016 with 2011 as the turning point, which shows the two-stage characteristics of stable development and fluctuation period. On the other hand, the gradient differentiation pattern of “ Weak East and Strong West” is significant spatially. However, the dual spatial pattern is gradually weakened since the innovation efficiency has been descending from the west to the east. (2) In terms of productivity of urban innovation activities, the Malmquist index increased from 6.2% in 2006 to 18.4% in 2016 indicates that the growth trend is relatively stable, while the generally unobvious spatial differentiation indicates the three-stage evolutionary characteristics of “divergence-convergence-divergence”. (3) GMM regression results show that the driving factors have different spatial effects on both of the whole urban agglomeration and the three sub-urban groups in the middle reaches of Yangtze River. The informationization process, the economic foundation, and the foreign business activity positively help to inspire urban innovation efficiency; the financial scale has a significantly negative effect; the tertiary industry scale, innovation platform of university and government support have heterogenous effects obviously. This study has the contribution to improve the regional innovation environment for the local governments.