This paper presents a novel low-complexity multiple-input multipleoutput (MIMO) detection scheme using a distributed M-algorithm (DM) to achieve high performance soft MIMO detection. To reduce the searching complexity, we build a MIMO trellis graph and split the searching operations among different nodes, where each node will apply the M-algorithm. Instead of keeping a global candidate list as the traditional detector does, this algorithm keeps multiple small candidate lists to generate soft information. Since the DM algorithm can achieve good BER performance with a small M, the sorting cost of the DM algorithm is lower than that of the conventional K-best MIMO algorithm. The proposed algorithm is very suitable for high speed parallel processing.