Atttribute reduction is a hot point in the machine learning and knowledge discover research, while the attribute importance measurement is the key link in the structure of the attribute reduction algorithm. For the imcomplete of the mixed information system, a new measurement method of attribute integration importance, named neighborhood combination measure, is defined under the neighborhood relation, and a neighborhood combination measure based attribute reduction(NCMAR) algorithm is also proposed. Some experiments are carried out on UCI data sets. And the experiments results show that the NCMAR algorithm can not only deal with mixed decision system with symbol data and numerical data, but is suitable for the imcomplete information system. What’s more, it can obtain smaller reducts and better classification accuracy than current algorithms.