Deep learning recently achieved impressive successes on various computer vision tasks for which large amounts of training data is available, such as image classification. These successes have motivated the use of computer graphics to generate synthetic data for tasks where real data is difficult to collect. We present SynDraw, a non-photorealistic rendering system designed to ease the generation of synthetic drawings to train data-driven sketch-based modeling systems. SynDraw processes triangular meshes and extracts various types of synthetic lines, including occluding contours, suggestive contours, creases, and demarcating curves. SynDraw exports these lines as vector graphics to allow subsequent stylization. Finally, SynDraw can also export attributes of the lines, such as their 3D coordinates and their types, which can serve as ground truth for depth prediction or line labeling tasks. We provide both a command- line interface for batch processing, as well as an interactive viewer to explore and save line extraction parameters. We will release SynDraw as an open source library to support research in non-photorealistic rendering and sketch-based modeling.