In this paper, a new blind equalization cost function, termed differential mean output energy (DMOE), was presented, and a new multiuser detection algorithm with variable step size was designed. It is shown through simulation results for a co-channel system involving severe MAI that the DMOE algorithm gives significantly enhanced signal to interference ratio (SIR) performance and converges rapidly to the optimum MMSE detector, together with a low computational complexity requirement relative to the recursive least squares (RLS) algorithm. Thus, it is not necessary to switch to the decision-directed mode. Moreover, it also exhibits global convergence, and can be used in different interference environment without the requirement of estimating and restricting the surplus energy.