Abstract: Case based reasoning (CBR) approach is based on knowledge acquisition and it is a new method for data driven modeling. The key stage of CBR system is the case retrieval one. This paper presents an improved modeling method of K nearest neighbor regression for case retrieval system. Firstly, a cluster based nearest neighbor algorithm is used to divide the case library for improving the quality of case retrieval. Secondly, for the problem of selecting the most suitable number of neighbors, this paper adopted the particle swarm algorithm, instead of the traditional empirical way. The simulation via Mackey Glass chaotic time series data validated the feasibility of the proposed method.