In this paper, we propose radar system recognition approach using a HMM. We employ the Baum-Welsh algorithm to search for an HMM which best explains the observed radar signals represented by sequences of 0's and 1's. Deterministic periodic sequences (stable PRI radars) are considered. We can obtain an HMM which yields the globally biggest training probability. We can modify the results, to some degree, to create models more robust to observation errors. Preliminary results in combination with either the forward back-ward procedure or the Viterbi algorithm may be adequate for carrying out radar system recognition.