This survey provides a comprehensive overview of the evaluation of Retrieval-Augmented Generation (RAG) systems. RAG integrates information retrieval with generative models to enhance the accuracy and reliability of language models. The paper discusses the challenges in evaluating RAG systems, including the complexity of the hybrid structure, the dynamic nature of external knowledge sources, and the need for effective metrics to assess retrieval and generation components. The authors propose a unified evaluation process (Auepora) to systematically analyze RAG benchmarks, focusing on three key aspects: evaluation targets, datasets, and metrics. The paper highlights the importance of evaluating retrieval accuracy, generation quality, and the overall system performance. It discusses various metrics used in RAG evaluation, such as precision, recall, and F1 scores, as well as more specialized metrics for assessing faithfulness, correctness, and diversity. The authors also emphasize the need for diverse and comprehensive datasets that reflect real-world scenarios and the challenges in aligning evaluation metrics with human preferences. The paper identifies key challenges in RAG evaluation, including the dynamic nature of information sources, the need for robustness against noise and ambiguity, and the importance of latency and response quality. It also discusses the role of large language models (LLMs) in evaluating RAG systems and the potential of using LLMs as evaluative judges. The authors suggest that future research should focus on developing more comprehensive and holistic evaluation frameworks that address the complexities of RAG systems and their practical applications. The survey concludes that a systematic and structured approach to evaluating RAG systems is essential for advancing the field and improving the effectiveness of these systems in real-world scenarios.This survey provides a comprehensive overview of the evaluation of Retrieval-Augmented Generation (RAG) systems. RAG integrates information retrieval with generative models to enhance the accuracy and reliability of language models. The paper discusses the challenges in evaluating RAG systems, including the complexity of the hybrid structure, the dynamic nature of external knowledge sources, and the need for effective metrics to assess retrieval and generation components. The authors propose a unified evaluation process (Auepora) to systematically analyze RAG benchmarks, focusing on three key aspects: evaluation targets, datasets, and metrics. The paper highlights the importance of evaluating retrieval accuracy, generation quality, and the overall system performance. It discusses various metrics used in RAG evaluation, such as precision, recall, and F1 scores, as well as more specialized metrics for assessing faithfulness, correctness, and diversity. The authors also emphasize the need for diverse and comprehensive datasets that reflect real-world scenarios and the challenges in aligning evaluation metrics with human preferences. The paper identifies key challenges in RAG evaluation, including the dynamic nature of information sources, the need for robustness against noise and ambiguity, and the importance of latency and response quality. It also discusses the role of large language models (LLMs) in evaluating RAG systems and the potential of using LLMs as evaluative judges. The authors suggest that future research should focus on developing more comprehensive and holistic evaluation frameworks that address the complexities of RAG systems and their practical applications. The survey concludes that a systematic and structured approach to evaluating RAG systems is essential for advancing the field and improving the effectiveness of these systems in real-world scenarios.