A major performance-limiting factor in terrestrial optical wireless (OW) systems is turbulence-induced fading. Exploiting the additional degrees of freedom in the spatial dimension, multiple laser transmitters combined with multiple receiver apertures provide an effective solution for fading mitigation. Although MIMO (multiple-input multiple-output) OW systems have been extensively studied in recent years, most of these works are mainly limited to symbol-by-symbol decoding. Maximum likelihood sequence detection (MLSD) exploits the temporal correlation of turbulence-induced fading and promises further performance gains. In this paper, we investigate MLSD for IM/DD (intensity-modulation/direct-detection) MIMO OW systems over log-normal atmospheric turbulence channels. Even with a low-order modulation scheme such as on-off keying which is typically used in OW systems, the complexity of MLSD might be prohibitive. We therefore present an iterative sequence detector based on the expectation-maximization (EM) algorithm. The complexity of the proposed algorithm is much smaller than a direct evaluation of the log-likelihood function. The Monte-Carlo simulation results demonstrate that the EM-based algorithm outperforms the symbol-by-symbol decoder and achieves a performance which lies within 0.5 dB of that of the optimal MLSD.