The paper introduces the CHAIN-OF-TABLE framework, which enhances table-based reasoning using large language models (LLMs). Unlike generic reasoning approaches, CHAIN-OF-TABLE explicitly incorporates tabular data into the reasoning chain as a proxy for intermediate thoughts. The framework guides LLMs to iteratively generate operations and update the table, forming a chain of tables that evolve to represent the reasoning process. This continuous evolution of the table helps LLMs dynamically plan the next operation based on the results of previous ones, leading to more accurate and reliable predictions. The method is evaluated on three benchmarks (WikiTQ, FeTaQA, and TabFact) using various LLMs (PaLM 2, GPT-3.5, and LLaMA 2), achieving state-of-the-art performance. The paper also discusses the effectiveness of different table operations and their impact on performance, as well as the efficiency and scalability of the proposed method.The paper introduces the CHAIN-OF-TABLE framework, which enhances table-based reasoning using large language models (LLMs). Unlike generic reasoning approaches, CHAIN-OF-TABLE explicitly incorporates tabular data into the reasoning chain as a proxy for intermediate thoughts. The framework guides LLMs to iteratively generate operations and update the table, forming a chain of tables that evolve to represent the reasoning process. This continuous evolution of the table helps LLMs dynamically plan the next operation based on the results of previous ones, leading to more accurate and reliable predictions. The method is evaluated on three benchmarks (WikiTQ, FeTaQA, and TabFact) using various LLMs (PaLM 2, GPT-3.5, and LLaMA 2), achieving state-of-the-art performance. The paper also discusses the effectiveness of different table operations and their impact on performance, as well as the efficiency and scalability of the proposed method.