This repository contains images of maize (corn) leaves that have been annotated to mark lesions caused by Northern Leaf Blight (NLB), a common and devastating disease of maize. In total, there are 18,222 images, all taken in the field, and 105,735 annotations by one of two human experts. This is the largest publicly available collection of classified images of any single plant disease. The central goal of this project is to develop machine learning algorithms capable of accurately diagnosing plant disease from images taken in the field. This would enable rapid image-based disease screening of large areas, a valuable technology for plant breeders, geneticists, and farmers. **Why annotate each lesion?** Because field images are very noisy, with many types and shapes of dead tissue visible, annotations showing the positions of lesions are much more valuable than classifications of the entire image. Work by our group (DeChant et al., *Phytopathology* 107) showed that images simply classified as "visible lesions(s)" or "no lesions" could only train a CNN to a low degree of accuracy, while training CNNs on images with annotated lesions gave a final accuracy of 96.7%. **Why experts?** Non-experts cannot accurately distinguish lesions caused by NLB from those caused by other diseases, insect feeding, etc. Thus we could not rely on non-experts (e.g. Mechanical Turk workers) to generate training data, as one might for easily-recognizable objects. Having a large amount of trustworthy, expert-generated data is a valuable resource. **Why NLB?** This disease, caused by the fungus *Setosphaeria turcica*, has been growing more severe in recent years. In 2015 it was the most economically damaging maize disease in the US and Ontario, causing roughly $2 billion in damage. Because the disease manifests as large, necrotic lesions, it is a tractable candidate for remote detection with images. A full description of the files is being prepared, to be submitted BMC Data Notes. This work was funded by the NSF National Robotics Initiative project #1527232, Deep Learning Unmanned Aircraft Systems for High-Throughput Agricultural Disease Phenotyping.