Abstract: Underwater very low frequency (VLF) minimum shift keying (MSK) signals are distorted by the impulsive noise, the propagation attenuation of sea water and the fluctuation of sea wave. In order to reduce the requirement for carrier phase synchronization and improve the robustness of demodulation algorithm, the differential demodulation algorithm of the underwater VLF MSK signal is investigated in this paper. The propagation attenuation of sea water is first modeled. By considering the impact of suppressing impulsive noise, the multiple-symbols differential demodulation of the underwater VLF MSK signal is then derived based on the principle of maximum likelihood (ML) detection. Extensive simulation results show that the performance of the ML multiple-symbols differential demodulation algorithm can closely approach that of the coherent demodulation algorithm if the carrier phase is recovered and a relative large amount of symbols are utilized. On the other hand, the differential demodulation algorithm is superior to the coherent demodulation algorithm if the carrier phase is not recovered.