The mining of cube gradient is a new research problem in the data mining community. It is an extension of the traditional association rules mining in the data cube and its purpose is to mine the changes of measures in a multidimensional space. However, previous works explore for cube gradients with raw number, and the semantics of the changes of cube gradients, such as high or low, are not clear and those are not easy to be understood for the decision makers. In this paper, the fuzzy gradient mining problem with fuzzy constraint sets is presented. The mining algorithm, MFG, for the problem in the case of materialized cube is presented. In MFG, the condensed cube technology is used to reduce the searching space of the mining algorithm. The performance tests based on the synthesized data sets show the mining algorithm is effective and scalable with the fuzzy constraint sets.