This study proposes a generative design method for tire tread patterns that incorporates latent design intent extracted from patent documents. As various tread patterns satisfy primary performance criteria resulting from different combinations of grooves, generative design is a suitable means for the early design stage. However, designers cannot control the exploration range due to the lack of correlation between geometric characteristics and evaluation criteria. While their correlation is often implicit in practice, certain aspects are described in patent documents to claim their originality. The proposed method aims to extract the correlation between geometric characteristics and evaluation criteria from patent documents and apply them to the generative design of tire tread patterns. Firstly, tread pattern images and their characteristic summaries are constructed from drawings and descriptions in patent documents related to a design problem. Then, the geometric features of the pattern images are extracted using image processing techniques from a reverse engineering perspective. Secondly, categories that share some characteristics are identified through a concept identification approach. Finally, generative design explores promising alternatives under geometric constraints derived from the selected category that satisfies the desired evaluation criteria. This study demonstrates the capability of the proposed method in the design problem of tread patterns for summer passenger car tires.