In this correspondence, we investigate the decision-feedback detection for multiple-input multiple-output intersymbol-interference (MIMO-ISI) channels. First, a novel constrained symbol-by-symbol decision-feedback detector (DFD) is proposed, in which a constraint on the feedback filter provides robustness against error propagation and outperforms the conventional DFD. However, we find that an error floor is observed at high signal-to-noise ratios (SNRs) if a fixed constraint is used. To resolve this problem, we then propose an iterative symbol-by-symbol DFD in which an adaptive constraint is implicitly used to update the DFD's coefficients. Simulation results show that the error-floor problem is overcome and that the performance becomes satisfactory at high SNRs through iterations.