Both spatial modulation (SM) and spatial multiple access (SMA) exploit spatial differences between antennas to implement modulation and multiple access. In this sense, this letter designs an SM-SMA system for ambient backscatter communication networks. SM activates a part of a tag’s antennas to carry information, and thus decreases the interference between antennas in SMA. We then propose a modified-maximum likelihood (M-ML) detector to separate and obtain and the backscattered signal without the need of recovering the source signal. By exploiting the sparse characteristics of the SM-SMA signal, a multi-user sparse Bayesian learning (MSBL) based detector is further proposed to reduce the complexity of M-ML. The simulation results verify that SM-SMA achieves a lower bit error rate than SMA, and the MSBL based detector has a lower complexity than M-ML at the cost of increasing BER.