In recent years, research groups pay even more attention on 3D images, especially in the field of biomedical image processing. Adding another dimension enables to capture the entire object. On the other hand, handling 3D images also requires new algorithms, since not all of them can be modified for higher dimensions intuitively. In this article, we introduce a comparison of various implementations of 3D texture descriptors presented in the literature in recent years. We prepared an unified environment to test all of them under the same conditions. From the results of our tests we came to conclusion, that 3D variants of LBP in the combination with k-NN classifier are a very strong approach with the classification accuracy more than 99% on selected group of 3D biomedical images.