Gait training is a major focus of rehabilitation for many people with neurological disorders, yet systematic reviews have failed to identify the most effective form of gait training. The main objective of this study was to compare conditions for gait training for people with acquired brain injury (ABI). Seventeen people who had sustained an ABI and were unable to walk without assistance were recruited as a sample. Each participant was exposed to seven alternative gait training conditions in a randomized order. These were: (1) therapist manual facilitation; (2) the use of a gait-assistive device; (3) unsupported treadmill walking; and (4) four variations of body weight support treadmill training (BWSTT). Quantitative gait analysis was performed and Gait Profile Scores (GPS) were generated for each participant to determine which condition most closely resembled normal walking. BWSTT without additional therapist or self-support of the upper limbs was associated with more severe gait abnormality [Wilks' lambda = 0.20, F(6, 6) = 3.99, p = 0.047]. With the exception of therapist facilitation, the gait training conditions that achieved the closest approximation of normal walking required self-support of the upper limbs. When participants held on to a stable handrail, self-selected gait speeds were up to three times higher than the speeds obtained for over-ground walking [Wilks' lambda = 0.17, F(6, 7) = 5.85, p < 0.05]. The provision of stable upper-limb support was associated with high self-selected gait speeds that were not sustained when walking over ground. BWSTT protocols may need to prioritize reduction in self-support of the upper limbs, instead of increasing treadmill speed and reducing body weight support, in order to improve training outcomes.