This paper proposes a novel approach (short for iEDA/TFAD) based on estimation of distribution algorithms and time-frequency atom decomposition for analyzing radar emitter signals. In iEDA/TFAD, an improved estimation of distribution algorithm combining Gaussian and Cauchy probability models is presented to implement time-frequency atom decomposition to analyze several typical radar emitter signals by extracting their features and recognizing them. The introduction of iEDA can greatly reduce the computational complexity of TFAD. Experimental results show that EDA/TFAD can efficiently recognize several radar emitter signals at a high correct rate.