This paper applies data envelopment analysis (DEA) and stochastic frontier analysis (SFA) to explore the relative efficiency of China's research universities of science and technology. According to the finding, when talent training is the only output, the efficiency of research universities of science and technology is far lower than that of comprehensive universities. However, when the outputs of scientific research and social service are taken into consideration, the former will increase remarkably and even exceeds the latter. Moreover, the research universities of science and technology have a higher inner homogeneity because their standard deviation of efficiency score is obviously lower. The study also reveals that the number of research universities of science and technology with redundant inputs is more than that of comprehensive universities. Beyond that, the environmental variables and statistical noise exert no significant effect on the efficiency of Chinese research universities.