We present three different decision-feedback sequence estimation (DFSE) schemes for time-reversal space-time block coding (TR-STBC). The first scheme is called unwhitened DFSE (U-DFSE) and performs reduced-state sequence estimation based on the output of the spatio-temporal matched filter (MF) typically employed in TR-STBC. The second approach improves upon U-DFSE by subtracting a bias term caused by anti-causal interference from the U-DFSE metric. In the third scheme, the noise component in the output of the spatio-temporal MF is first whitened using a prediction-error filter that can he efficiently computed using the Levinson-Durbin algorithm, and subsequently whitened DFSE (W-DFSE) is performed. As relevant example, all three DFSE schemes are compared for the GSM/EDGE system and typical channel profiles such as typical urban (TU) and hilly terrain (HT). Our results show that for binary modulation (as used in GSM) U-DFSE and its improved version can approach the performance of W-DFSE for the full range of delay spreads relevant for GSM and EDGE. On the other hand, for high-level modulation (as used in EDGE) only W-DFSE gives a satisfactory performance, if a low trellis complexity is desired.