There is a large number of incomplete continuous data in real life and classical rough set theory is only available for discrete data. In such a case, a new neighborhood rough set model for incomplete information system is proposed to classify the objects by using the distance set of attribute value. It is proved that neighborhood relation is equivalent to the similarity relation when threshold value is zero, and neighborhood relation also has the same meaning with indiscernibility relation on the condition that the information system is complete and threshold value is zero. A heuristic knowledge reduct algorithm based on neighborhood relation is provided and an example of knowledge reduct to several different information systems confirms the effectiveness of neighborhood rough set model for incomplete information system.