This article presents a dataset of breast ultrasound images for research purposes. The dataset includes 780 images of breast ultrasound scans, categorized into three classes: normal, benign, and malignant. The images are in PNG format, with an average size of 500×500 pixels. The dataset was collected from 600 female patients aged between 25 and 75 years, and was collected in 2018. The images were collected from Baheya hospital using a LOGIQ E9 ultrasound system and LOGIQ E9 Agile ultrasound system. The images were then preprocessed to remove unnecessary information and to improve the quality of the images. The preprocessing steps included removing duplicate images, converting DICOM images to PNG format, and cropping the images to remove unimportant boundaries. The dataset also includes ground truth images, which are used to evaluate the performance of machine learning models. The ground truth images are created using a freehand segmentation method in Matlab. The dataset is available for researchers to use in their studies on breast cancer classification, detection, and segmentation. The dataset is the first publicly available breast ultrasound dataset. The authors of the article have declared no conflict of interest. The dataset is available online at https://doi.org/10.1016/j.dib.2019.104863.This article presents a dataset of breast ultrasound images for research purposes. The dataset includes 780 images of breast ultrasound scans, categorized into three classes: normal, benign, and malignant. The images are in PNG format, with an average size of 500×500 pixels. The dataset was collected from 600 female patients aged between 25 and 75 years, and was collected in 2018. The images were collected from Baheya hospital using a LOGIQ E9 ultrasound system and LOGIQ E9 Agile ultrasound system. The images were then preprocessed to remove unnecessary information and to improve the quality of the images. The preprocessing steps included removing duplicate images, converting DICOM images to PNG format, and cropping the images to remove unimportant boundaries. The dataset also includes ground truth images, which are used to evaluate the performance of machine learning models. The ground truth images are created using a freehand segmentation method in Matlab. The dataset is available for researchers to use in their studies on breast cancer classification, detection, and segmentation. The dataset is the first publicly available breast ultrasound dataset. The authors of the article have declared no conflict of interest. The dataset is available online at https://doi.org/10.1016/j.dib.2019.104863.