Abstract: In radar countermeasure and counter-countermeasure, radar and jammer should mutually sense, identify, recognize, and counter each other. To solve the problem of non-cooperative system identification and inference for the radar side to the jamming system side, an inverse inference method was proposed to analyze the results of radar emitter work mode in adversary jammers. Firstly, an inverse recognition model was established to recognize radar emitter work mode, wherein the jamming actions were optimized based on the recognized radar work mode. In radar’s side (our side), the jammer’s recognition result of the radar work mode was estimated based on the observed jamming action sequence, and then an inverse inference method was designed based on hidden Markov models (HMM). Specifically, the adaptive particle swarm optimization algorithm was used to initialize the HMM parameters, then the HMM parameters were estimated through the Baum-Welch algorithm with multiple observation sequences, and the jammer’s radar work mode recognition results were estimated through the logarithmic Viterbi algorithm. Finally, the feasibility and effectiveness of the proposed inverse analysis method were proved under typical radar countermeasure scenarios.