This paper investigates the decision feedback detection for intersymbol interference multiple-input multiple-output (MIMO-ISI) channels. Firstly, a constrained symbol-by-symbol decision feedback detector (CS-DFD) is presented, in which a fixed constraint is used to mitigate the error propagation. Then, we propose an iterative symbol-by-symbol decision feedback detector (IS-DFD) which possibly minimizes the amount of decision error through iterations. While both methods outperform the conventional decision feedback detection, the IS-DFD can overcome the error floor problem from which the C-DFD suffers at a high signal-to-noise ratio.