Artificial intelligence (AI) may aid in triaging skin diseases. However, most AI models have not been rigorously assessed on images of diverse skin tones or uncommon diseases. To ascertain potential biases in algorithm performance in this context, we curated the Diverse Dermatology Images (DDI) dataset—the first publicly available, deeply curated, and pathologically confirmed image dataset with diverse skin tones. The DDI was retrospectively selected from reviewing pathology reports in Stanford Clinics from 2010-2020 with further details provided in the methods. After filtering out images that were poor quality (see methods), there were a total of 656 images representing 570 unique patients (Supplemental Table 2 and 3). The dataset comprised a retrospective convenience sample across all images of FST I-VI but was also designed to allow direct comparison between FST I-II and FST V-VI by matching diagnostic category, age within 10 years, gender, and date of photograph within 3 years (online methods). There was no significant difference in photo quality scores between FST I-II photos and FST V-VI photos (Mann-Whitney U, p = 0.33). There were a total of 208 images of FST I-II (159 benign, 49 malignant), 241 images of FST III-IV (167 benign, 74 malignant), and 207 images of FST V-VI (159 benign and 48 malignant).