Abstract: Radar operational mode recognition is achieved through the analysis of intercepted signals to identify radar functions and behavioral states, which is an important part of electronic reconnaissance and electronic countermeasures. The development of knowledge-driven and data-driven algorithms is examined from different contextual backgrounds. The existing algorithms are summarized in four stages: statistical analysis, behavioral reasoning, traditional machine learning, and deep learning, highlighting their fundamental concepts, innovations, and limitations. Additionally, the current challenges and difficulties in the research are discussed. Statistical analysis methods effectively address recognition issues under clear signal backgrounds and simple modulation types. Behavioral reasoning methods utilize probabilistic calculations to analyze the intrinsic relationships among radar modes, demonstrating capabilities in conventional pattern recognition. Traditional machine learning extracts deeper patterns from data distributions, enabling pattern recognition under complex conditions with human assistance. In contrast, deep learning methods largely eliminate human intervention, achieving automated processing through end-to-end recognition.