Abstract. An algorithm for solving the "learning parity with noise" (LPN) problem is proposed and analyzed. The algorithm originates from the recently proposed advanced fast correlation attacks, and it employs the concepts of decimation, linear combining, hypothesizing and minimum distance decoding. However, as opposed to fast correlation attacks, no preprocessing phase is allowed for the LPN problem. The proposed algorithm appears as more powerful than the best one previously reported known as the BKW algorithm proposed by Blum, Kalai and Wasserman. In fact the BKW algorithm is shown to be a special instance of the proposed algorithm, but without optimized parameters. An improved security evaluation, assuming the passive attacks, of Hopper and Blum HB and HB + protocols for radio-frequency identification (RFID) authentication is then developed. Employing the proposed algorithm, the security of the HB protocols is reevaluated, implying that the previously reported security margins appear as overestimated.