Fuzzy decision tree(FDT) is an expansion and improvement of decision tree, which expands the application scope of decision tree to deal with the uncertainty and extracts classification rules efficiently. However, there are some problems in the process of sample attributes fuzzifying, such as information loss and the influence of strong personal subjectivity on the results. To solve these problems, a fuzzy decision tree algorithm based on Z-number(Z-FDT) is put forward. we propose a method to construct Z-number, first, membership functions is defined to fuzzify the data set and describe the size of the probability measure of the fuzzy set, then the cluster centers of multi-dimensional data are extracted ,using the fuzzy clustering algorithm., the membership function and cluster centers of different categories fuzzy partition the sample data set based on the features of the sample, and obtain the fuzzy sets. Finally, based on the definition of Z-number, fuzzy sets are constructed into Z-numbers. After the multidimensional data set is transformed into Z-numbers, the uncertainty measure of Z-number is used to select appropriate features to partition the data set, until the stop condition is met. Z-FDT can directly classify and predict samples with continuous attribute values, and has good generalization ability and relatively stable results, at the last, this will be confirmed through the comparative experiment with ID3, Cart and Fuzzy ID3.