Based on the idea of neighborhood granulation

the rough set model for fuzzy decision information system was proposed.The upper and lower approximation operators were defined and their properties were discussed.By constructing the discernibility matrix and discernibility function

a new method of upper and lower consistent attribute reductions was given for a fuzzy decision information system with continuous values of condition attributes.The method was verified by an illustrative example.It provides a new approach to attribute reduction and feature selection in fuzzy decision information systems.