We investigate soft-input Turbo channel estimation for multiple-input-multiple-output (MIMO) systems with single-carrier (SC) frequency-domain equalization (FDE), based on the Kalman filtering and normalized recursive least-squares (NRLS) criteria. Under each criterion, structured channel estimation (SCE) is developed, which exploits the correlation between adjacent frequency bins, and achieves a close performance to the case with perfect channel state information (CSI). An intensive performance analysis is provided in terms of the mean square error (MSE) of channel estimation, which provides a good match with the numerical results. Given a practical channel model for wireless communications, Turbo NRLS SCE (TNRLS-SCE) outperforms Turbo Kalman SCE (TK-SCE) in terms of the steady-state MSE, with a wide range of root-mean-square (RMS) delay spreads and a medium-to-high Doppler spread.