A new theoretical framework is introduced for analyzing the performance of a finite length minimum-mean-square error decision feedback equalizer (MMSE-DFE) in a multi-input multi-output (MIMO) environment. The framework includes transmit and receive diversity systems as special cases and quantifies the diversity performance improvement as a function of the number of transmit/receive antennas and equalizer taps. Fast and parallelizable algorithms for computing the finite-length MIMO MMSE-DFE are presented for three common multi-user detection scenarios.