In this study, low-complexity sphere decoding (SD) algorithms for generalized spatial modulation (GSM) are proposed. To reduce the computational complexity of the SD algorithm for GSM, layer ordering and transmit antenna combination (TAC) partitioning are employed. The layer ordering scheme determines the detection order of layers based on their reliabilities, which are measured by their minimum mean-square errors (MMSEs). In TAC partitioning, which is combined with the early termination technique, the likely TACs are selected by exploiting an MMSE-based weighting metric to reduce the search space. Simulation results demonstrate that the proposed ordering and partitioning aided SD schemes achieve nearly the same performance as the conventional SD scheme; however, they require significantly low complexity, especially in large GSM multiple-input-multiple-output systems and in the low SNR region.