The statistically optimum receivers give a good performance in binary data transmission systems subject to severe intersymbol interference and statistical noise, but they become too complex to be practical. This paper describes the new practical detector which can be derived by introducing the idea of the decision feedback to the maximum likelihood (ML) bit estimation, and which retains the implementable simplicity like conventional decision feedback equalizers (DFE). The proposed detector has the multilayered structure as if the multiple conventional DFE's are arranged in a parallel form. A method is also developed to estimate the error probability of the detector, and its analysis and simulation results show that, by increasing the level of the multilayers and the tap number of the feedforward filter, the proposed scheme can ahcieve the error rate performance very near to that of the ML bit estimation.