In this paper, we develop and present a DFT-SOFDM system (operating in both distributed and localized FDMA modes) for transmission and reception through a frequency-selective, slow fading channel. This system allows us to use well-known linear receiver techniques such as the zero-forcing (ZF) and minimum mean-squared error (MMSE) equalization. More importantly, we proposed using a sphere decoding algorithm (SDA) on DFT-SOFDM to obtain a maximum-likelihood (ML) estimate on the received signal, hence achieving superior bit error rate (BER) performance without the complexity of an exhaustive search. Simulation results show that SDA gives superior BER performance for both distributed and localized DFT-SOFDM compared to suboptimal ZF or MMSE receivers.