Abstract: In practical evaluation, people usually prefer to select a set of indicators or establish indicator system with hierar chical structure to present a relative comprehensive evaluation, and the importance of indicators in different level is of ten different.However, it is hard for standard DEA models to handle too many inputs and outputs indicators directly because of discrimination power.Meanwhile, standard DEA models also cannot handle hierarchical indicators directly.Based on the preference and measurement implied in DEA models, this study presents multi-level DEA models, which can deal with the indicators with hierarchical structure directly.Additionally, value judgments of the decision makers are well incorporated, and discrimination power of DEA model is much enhanced when many indicators are used.A practical evaluation of 15 institutes for basic research institutions is provided at the end.

Key words: DEA, discrimination power, multi-level DEA models