A major challenge in tumor surgery is the differentiation between normal and malignant tissue. Since an incompletely resected tumor easily leads to recidivism, the gold standard is to remove malignant tissue with a sufficient safety margin and send it to pathology for examination with patho-histological techniques (rapid section diagnosis). This approach, however, exhibits several disadvantages: The removal of additional tissue (safety margin) means additional stress to the patient; the correct interpretation of proper tumor excision relies on the pathologist’s experience and the waiting time between resection and pathological result can be more than 45 minutes. This last aspect implies unnecessary occupation of cost-intensive operating room staff as well as longer anesthesia for the patient. Various research groups state that hyperspectral imaging in the mid-infrared, especially in the so called "fingerprint region", allows spatially resolved discrimination between normal and malignant tissue. All these experiments, though, took place in a laboratory environment and were conducted on dried, ex vivo tissue and on a microscopic scale. It is therefore our aim to develop a system incorporating the following properties: Intraoperatively and in vivo applicable, measurement time shorter than one minute, based on mid infrared spectroscopy, providing both spectral and spatial information and no use of external fluorescence markers. Theoretical assessment of different concepts and experimental studies show that a setup based on a tunable Quantum Cascade Laser and Attenuated Total Reflection seems feasible for in vivo tissue discrimination via imaging. This is confirmed by experiments with a first demonstrator.