An adaptive multiuser detection (MUD) technique using the complex radial basis function (CRBF) network is proposed for space division multiple access-orthogonal frequency division multiplexing (SDMA-OFDM) system. Among various MUDs, the linear minimum mean-square error (MMSE) MUD suffers from poor performance and the maximum likelihood (ML) detector is restricted by high computational complexity. Hence, the cost function minimisation-based detector like minimum symbol error rate (MSER) is preferred because of significant performance gain over MMSE MUD and complexity gain over ML detector. Moreover, the MSER detector also has a potential of surviving in overload scenario, where the number of users are more than that of the number of receiving antennas. However, in all these techniques, the requirement of channel estimation adds an extra complexity whereas, the proposed CRBF detector approximates the channel parameters in training phase and detects signals in testing phase. It also has low complexity, better performance compared with MSER MUD and also supports overload scenario. Each neuron in the proposed CRBF network is assembled with `sech' activation function, as this function can do better complex non-linear mapping than Gaussian activation. The simulation study and performance evaluation of CRBF MUD is investigated, considering both data and image transmission.