Formal concept analysis and rough set theory are two effective data analysis methods, which merge with each other. Through introducing fuzzy rough sets into fuzzy formal concept analysis, based on generalized fuzzy rough approximations, a crisp-fuzzy concept lattice is proposed. Subsequently, the attribute reduction of the corresponding concept lattice is studied. The study involves the definition of attribute reduction to keep the lattice structure of the concept lattice unchanged, the judgement of consistent attribute sets, the feature characterization of different types of attributes, and a method of reduction computation based on discernibility matrix.