We seek to test whether innovation, measured by the number of accepted patents improves or worsens the environment in China. We hypothesize the existence of an inverse U-shaped curve, which differs by industry and provincial level of development. In that context, we test for a quadratic relationship between accepted patents and CO2 emissions per capita in 30 provinces and 32 economic sectors of China. We use a novel fixed effect panel data quantile (FEQR) regression estimator and differentiate between energy-intensive and non-energy intensive sectors, as well as between more and less-developed provinces of China. We find evidence for an inverse U-shaped relation between patent generation and CO2 emissions for both, more and less energy-intensive sectors, suggesting that at low levels of innovation new technologies tend to be “dirty”, but at high levels of innovation new technologies tend to be “green”. The same relationship holds for less-developed provinces as well. For more-developed provinces, we find the opposite relation, which we explain with a “rebound effect”.