近年来，研发效率问题成为学者关注的焦点，但还没有针对产业集群的研发效率的研究。本文选取研发经费内部支出和研发人员数量为产业集群研发效率的投入指标，以发明专利授权数为产出指标。并基于现有三阶段DEA未考虑决策单元之间合作关系的局限性，提出三阶段仁慈型DEA模型，并计算了2008年到2012年我国10个高新技术产业集群的研发效率，结果证实了方法的适用性。研究结果表明：（1）在剔除了环境和随机因素的影响后，各个高新技术产业的研发效率有明显的上升；（2）所在城市人均GDP和研究生毕业生数对研发经费内部支出的投入冗余和研发人员数量的投入冗余是正向有利因素，而政府资助研发经费占城市总研发经费的比重对研发经费内部支出的投入冗余是正向有利因素，对研发人员数量的投入冗余是负向不利因素。

In recent years, R&D efficiency has aroused wide concern among researchers, but few studies have involved the R&D efficiency of industry clusters. This paper selects internal R&D expenditure and number of R&D personnel as an input, and number of invention patents authorization is chosen as an output. Then this paper proposes the three-stage benevolent DEA model considering cooperation between decision-making units and calculates the R&D efficiencies of 10 industrial clusters during 2008-2012. The results verify the feasibility of the model and show that: (1) R&D efficiencies have increased obviously after the environmental factors and random factors are eliminated; (2) urban per capita GDP and number of graduates have a significant positive impact on the input redundancy of internal R&D expenditure and number of R&D personnel. The proportion of government R&D funds in urban total R&D expenditure has a positive influence in the input redundancy of internal expenditure of R&D while it has a negative influence in the number of R&D personnel.