The problem of optimally detecting coded data transmitted over channels which produce intersymbol interference is examined. The optimum bit-by-bit detector structure which directly estimates the transmitted data symbols from the received signal at the channel output is given. The performance of this joint detector is contrasted to that of a receiver in which the decision process is partitioned into the usual detector followed by a decoder. Even when the detector and decoder are separately optimized (bit-by-bit criterion), the joint detector is shown to considerably out perform the separately optimized structure, thereby illustrating the considerable loss of information resulting from partitioning of the decision process. Two near optimum algorithms which significantly reduce the complexity of the optimum joint detector are given. These relatively simple algorithms, which approximate the joint detector, are shown to be capable of performance superior to that of the more complex, separately optimized detector/decoder structure. The joint detector and the approximate algorithms presented offer an attractive alternative to the soft decision decoder which also attempts to overcome the loss of information resulting from sharp partitioning of the detector/ decoder decisions.