This dataset accompanies the journal article "Classification of Grapevine Varieties Using UAV Hyperspectral Imaging," published in Remote Sensing (MDPI): https://doi.org/10.3390/rs16122103. It contains hyperspectral images acquired by an unmanned aerial vehicle (UAV) covering seventeen grapevine varieties, grouped into red and white cultivars. The images are provided in their original, non-rectified form to reduce memory usage. Note that training was conducted on these unrectified images; the orthorectified mosaic for red varieties alone exceeds 70 GB and is not included due to size limits in Zenodo. Each hyperspectral image is accompanied by a header file (.hdr) and can be viewed and processed using SpectralView, which we highly recommend for both visualization and analysis. To facilitate quick inspection, an RGB extract is also provided for each hyperspectral image. Labeling was performed using the open-source tool Sensarea. NDVI images were generated and binarized to aid in segmentation. Rows of vines were then manually labeled within Sensarea based on these NDVI-derived masks. Code for loading and training on this dataset is available on GitHub.