We present an adaptive iterative (turbo) decision feedback equalizer (DFE) for channels with intersymbol interference (ISI). The filter coefficients are computed directly from the soft decisions and the received data to minimize a least squares cost function. This method performs better than a turbo DFE with perfect channel knowledge, where the filters are selected to minimize mean squared error (MSE) assuming perfect feedback. We also present adaptive reduced-rank algorithms, which can further improve performance and reduce complexity.