Based on the panel data of industrial enterprises in 28 provinces and cities in China from 2013 to 2017, the non-directed ERM model was used to measure the efficiency of innovation resource allocation and indicator invalidity in industrial enterprises, then the kernel density estimation and K-W test were used to make a systematic analysis. The results show that the allocation of innovation resources in China's industrial enterprises is not efficient on the whole and the distribution of efficiency within regions is significantly different, and the marginal returns of innovation factors such as labor and expenditure of industrial enterprises in multiple regions are decreasing. After 2015, some areas with low efficiency gradually shifted to high efficiency. The study also found that the surplus degree in the expenditure input on the technological transfer is serious, and, meanwhile, there is much shortage of potential outputs in the proportion of new product export. Therefore, improving the industrial innovation resource management system, advancing the application of new theories and technologies such as "Internet +", artificial intelligence, and accelerating the construction of a cross-region collaborative innovation platform are worthy paths to improve efficiency.