Buffer of Thoughts (BoT) is a novel thought-augmentation reasoning method designed to enhance the accuracy, efficiency, and robustness of large language models (LLMs). It introduces a meta-buffer to store high-level thought templates extracted from problem-solving processes across various tasks, known as thought templates. For each problem, a relevant thought template is retrieved and adaptively instantiated into a specific reasoning structure for efficient reasoning. Furthermore, a buffer manager is proposed to dynamically update the meta-buffer, thus augmenting its capacity as more tasks are solved.