In medicine, knowledge of sampled tissue origin and characteristic features of healthy organs is very important. Such data could lead to better abnormality detection and a smaller chance of laboratory mistakes. Although in most cases such information is gathered during the sampling process, sometimes there is a possibility of mislabeling by human error. This issue is especially apparent in scientific research, where very often such data are labeled by students or technicians who could lack the specialized knowledge from this field. Some of the time it is possible to determine the correct origin of the tissue by finding characteristics of the organ in the sample; however, often there is a lack of such data or the differences between abstract classes are just too small to do it even by a skilled professional. Such problems are apparent, for example, in the case of the same tissue but different animals. Another problem is the time requirements for laboratory examinations, where even if such validation could be made, it is impossible to meet the deadline. Because of that, an automated system able to perform highly accurate classification in a fraction of a second could be an important addition to such a laboratory. This paper presents a new dataset for healthy, multi-species tissue classification using light microscope imagery, containing 42 classes of abstraction. During the sampling process, over 7000 images were collected. Additionally, a custom, transformer-based, deep learning architecture was created that is able to classify between those classes with a high validation accuracy of over 96.6%. Such performance in some cases can be higher than that of a human specialist, especially in cases where characteristic features are not present in the sample. What’s more, such a dataset and deep neural network solution is one of the first of this kind in the veterinary medicine field, as most other, State-of-the-Art papers focus on human medicine or illness classification.