The adaptive differential detection (adaptive DD) of M-ary DPSK signals is proposed in which the phase reference for differential detection is estimated using the least mean square algorithm with step size optimally adapted to changing channel conditions. The Viterbi algorithm and decision feedback algorithm are applied to effectively perform the maximum likelihood detection of the transmitted phase sequence. The bit error rate (BER) performance of adaptive DD is evaluated by computer simulations of 4DPSK in additive Gaussian noise (AWGN) and Rayleigh fading channels. In AWGN channels, BER performance close to that of coherent detection with differential decoding is achieved with a performance loss of only 0.3 dB. In Rayleigh fading, the irreducible BER owing to fading Doppler spread can be reduced almost by half compared with conventional DD without prior knowledge of fading statistics.