Without considering differences in operating environment, traditional methods of efficiency evaluation are suffering from external environmental influences. This study presents an alternative approach for assessing the relative efficiency of government-sponsored research and development projects (GSP). A three-stage approach employing data envelopment analysis to evaluate efficiency and Tobit regression to control external variables was applied to 110 projects over 9 years. This study finds that firm size, industry, and ratio of public subsidy on research and development (R&D) budget of recipient firm significantly influences the technical efficiency of GSP in Taiwan. After controlling these external variables, the mean value of technical efficiency in the third stage increases and becomes significantly different to that in the first stage. Most GSPs increase their returns when their projects are scaled up. Furthermore, government policy makers must establish the upper-limit ratio of subsidies on R&D budgets of recipient firms to avoid inefficient use of public funds.