Protein classification is among the main themes in bioinformatics, for the reason that it helps understand the protein molecules. By classifying the protein structures, the evolutionary relations between them can be discovered. The knowledge for protein structures and the functions that they might have could be used to regulate the processes in organisms, which is made by developing medications for different diseases. In the literature, plethora of methods for protein classification are offered, including manual, automatic or semiautomatic methods. The manual methods are considered as precise, but their main problem is that they are time consuming, hence by using them a large number of protein structures stay uncategorized. Therefore, the researchers intensively work on developing methods that would afford classification of protein structures in automatic way with acceptable precision. In this paper, we propose an approach for classifying protein structures. Our protein voxel-based descriptor is used to describe the features of protein structures. For classification of unclassified protein structures, we use a k nearest neighbors classifier based on fuzzy logic. For evaluation, we use knowledge for the classification of protein structures in the SCOP database. We provide some results from the evaluation of our approach. The results show that the proposed approach provide accurate classification of protein structures with reasonable speed.