We optimize delay diversity (DD) for the case when computationally efficient decision-feedback equalization (DFE) is used at the receiver. We consider the general case of transmission over a correlated multiple-input multiple-output (MIMO) frequency-selective fading channel. The proposed optimization requires knowledge of the statistical properties of the wireless channel at the transmitter, but channel state information is only required at the receiver side. Based on an approximation of the pairwise error probability for DFE, we derive a stochastic gradient algorithm that allows us to optimize the DD transmit filters. Simulation results for the GSM/EDGE system show significant performance gains of the proposed optimized DD scheme over the DD schemes reported by D. Gore et al. (see Proc. IEEE Int. Conf. Commun., p.1949-53, 2002) and R. Schober et al. (see Proc. IEEE ICC, 2004), respectively, if DFE is used at the receiver.