For solid tumors, such as breast cancer, surgery is usually the treatment of choice. For these operations, it is of utmost importance to remove the whole tumor, since tumor-positive resection margins may result in recurrent disease and impaired overall survival. Current optical imaging techniques using endogenous contrast - e.g. diffuse reflectance spectroscopy or optical coherence tomography – for detection of breast cancers are limited to point measurements or long acquisition times. Broadband hyperspectral cameras, which provide a complete spectral fingerprint of the object at pixel level, are needed for two-dimensional imaging of the operative region. An <i>ex vivo</i> study was conducted to evaluate the feasibility of hyperspectral imaging for breast cancer detection. Fresh (<3 hours after surgery) resected breast cancer slices were imaged with a snapshot hyperspectral camera, which has 41 spectral bands, equally distributed in the visible and near-infrared (VIS-NIR) range (450 – 950 nm). Supervised analysis was performed by using the pathology annotations and unsupervised analysis was performed by using the hierarchical stochastic neighbour embedding (h-SNE) algorithm. So far, nine resected specimens, of which six invasive carcinomas and two (partially) mucinous carcinomas, were imaged. Spectral differences were found between the non-tumor, malign and benign regions on the resected specimens. Furthermore, automatic feature classification using h-SNE was possible in selected cases. Hyperspectral imaging showed great potential for discrimination of benign and malign breast tissue by using specific wavelengths bands in an <i>ex vivo</i> setting. Further analysis will be performed to determine whether it is possible to select tumor-specific wavelengths.