Generalization to previously unseen real-world data is the most important property of any machine learning algorithms. Yet, it is also one of the most challenging ones, as the data distribution used during training of, especially, deep learning algorithms, is commonly not representative of the real world use case. Domain generalization approaches try to mitigate this by applying tailored augmentation, sampling, or regularization techniques that encourage the model to learn more robust and transferable features. One area, where domain generalization is a particular issue, is the field of digital pathology, where the digital image is influenced by a significant number of parameters influencing the tissue preprocessing, staining procedure, and digitization. Moreover, the biological variability in tissue is considerable, leading to strong data variance depending on the tissue or tumor type, but also on the question which part of tissue is actually annotated and presented to the algorithms. The MIDOG 2025 challenge is providing yet another step towards the assessment of generalization for one of the most commonly tackled tasks in tumor pathology: Mitosis identification. The process of cell division (mitosis) is indicative of tumor growth and proliferation, and as such is part of the manual assessment of tumor alignancyby experts for many tumor types. Following up on the successful MIDOG 2021 and 2022 MICCAI challenges, we propose a third iteration of our challenge, focusing even more strongly on data variation that algorithms working in clinical environments must cope with. Previous challenges were limited towards pre-selected hot-spot regions, leaving the question of generalization to whole slide images open. This year, aiming to generalize to whole slide image applicability, the algorithms have to additionally show their robustness on a wider range of tissues, including non-hotspot regions or even non-tumor areas as well as inflammatory or necrotic areas, which occur in tumor samples regularly. Besides the main task of mitosis detection, another task of high scientific interest is covered in the challenge: mitotic figures with atypical morphologies have been shown to be an independent prognosticator of tumor malignancy. We therefore include the classification of normal vs. atypical mitotic figures as a second challenge track in the MIDOG 2025 challenge.