Currently, most of the methods for attribute reduction in formal decision contexts only consider single-granularity decision attributes, rarely investigating multi-granularity decision attributes. Motivated by this, this paper presents a method for attribute reduction in generalized multi-granularity formal decision contexts to achieve attribute reduction in information systems. Specifically, it first selects granularity for multi-granularity attributes and then removes redundant class attribute blocks. In order to characterize redundant class attribute blocks, this paper introduces conditional information entropy in generalized multi-granularity formal decision contexts, thereby distinguishing the importance of attributes to the information system, and subsequently proposes an attribute reduction method as well as its corresponding implementation algorithm. In the experiments, the data are first reduced using our method, and then the reduced data are used for achieving classification tasks. Our method is further compared with other existing algorithms in terms of the classification accuracy. The obtained results validate the effectiveness of the proposed attribute reduction method.