The AI4Agriculture Grape Dataset is a collection of 250 images taken in a vineyard in Ribera de Duero. The grapefruits are from the Tempranillo variety. The images are annotated using bounding boxes. The goal of this dataset is to provide images and the associated annotations to train and validate object detection models. These models can be used in viticulture applications, such as yield estimation, fruit counting, and field robotics. The images were taken in twelve different parcels. Six areas were defined in each parcel. Each area contains several lines of plants (grapevines). For each line of plants, 4 picture samples have been taken (two on each side). The samples are from non-contiguous grapevines, so the pictures never overlap. The pictures were taken by a Xiaomi Redmi 8 (with resolution 4032x3024), Xiaomi Mi A3 (with resolution 3000x4000) and a Xiaomi Redmi 9 (with resolution 3264x2448). To avoid showing grapes from other rows, the pictures were taken at approximately half the height of the vine-tree rows. For each image, a .xml file was generated, with the Pascal VOC format, containing all the labeled items (with bounding boxes) in that image. Annotations in YOLO format are also provided (*.txt). The labeling methodology is described in this document. Some images have issues: i) reflections caused by the sun ii) group of grapes partially hidden by leaves or iii) hidden by other objects (e.g. irrigation pipes), iv) color and direction similar to a tree trunk v) dead grapes in the floor or vi) grapes from the plant behind the plant analyzed. The images were captured by SmartRural and annotated by ATOS, UPC, Deutsches Zentrum f��r Luft- und Raumfahrt, and National and Kapodistrian University of Athens in the framework of the AI4Agriculture pilot of the AI4EU project.