Gene expressions measured using microarrays usually suffer from the missing value problem. Existing missing value imputation algorithms have some limitations. For example, some algorithms have good performance only when strong local correlation exists in data while some provide the best estimate when data is dominated by a global structure. In addition, these algorithms do not take into account many biological constraints in the imputation procedure. In this paper, we propose a set theoretic framework for missing data imputation. We design our algorithm by taking into consideration the biological characteristic of the data and exploit the local correlation and the global correlation structure adaptively. Experiments show that our algorithm can achieve a significant reduction of error compared with existing methods.