Abstract: With the development of the practice of local government purchasing criminal legal aid services, “how is the purchasing effectiveness” and “how to guarantee the purchasing effectiveness” become the key issues to be solved urgently. However, due to the fact that current evaluation modes are still limited to post-evaluation of services quality of contractors under the direct government provision mode, it is difficult to realize the complete evaluation of the performance of local government purchasing criminal legal aid service under the purchase mechanism. With the use of grounded theory and Delphi method, this paper, starting from the whole process of local government purchasing, with the consideration of the performance and the perception of that performance of three key actors—services buyer, services supplier, and services receiver – under multiple principal-agent relationships, builds a comprehensive evaluation model that is suitable for the evaluation of criminal legal aid services under the mode of government purchasing. The analysis of survey data through FCE and TOPSIS methods shows that: a) the overall performance of local government is relatively high but limited, showing the characteristics of “standard ‘basic actions’ and ‘to-be-strengthened’ ‘optional actions’”; b) government purchasing can achieve relatively high service process performance, service responsiveness and social benefits, but limited service satisfaction; c) relatively low performance of the local government is the main weak link; d) “sufficient competition” is not a necessary condition for the achievement of the performance. At last, efforts to improve purchase plan and make purchase price in line with the market price are the primary paths for performance improvement.