Postural instability and balance impairment are disabling symptoms in patients with acute unilateral peripheral vestibular hypofunction (UVH). Vestibular rehabilitation (VR) is known to improve the vestibular compensation process, but (1) its effect on posture recovery remains poorly understood, (2) little is known about when VR must be done, and (3) whether the degree of vestibular loss matters is uncertain. We analyzed posture control under static (stable support) and dynamic (unstable support) postural tasks performed in different visual conditions [eye open (EO); eyes closed (EC); and optokinetic stimulation] using dynamic posturography. Non-linear analyses of the postural performance (wavelet transform, diffusion analysis, and fractal analysis) were performed in two groups of patients with UVH subjected to the same VR program based on the unidirectional rotation paradigm and performed either early (first 2 weeks) or later (fifth to the sixth week) after vertigo attack. Distribution of the angular horizontal vestibulo-ocular reflex (aVOR) gain values recorded on the hypofunction side before rehabilitation differentiated two distinct sub-groups (cluster analysis) with aVOR gains below or above 0.20. The postural performance of the four sub-groups of patients with UVH (early rehabilitation with aVOR gain <0.20: n = 25 or gain >0.20: n = 19; late rehabilitation with aVOR gain <0.20: n = 15 or gain >0.20: n = 10) tested before VR showed significantly altered postural parameters compared with healthy controls. Greater instability, higher energy to control posture, larger sway without feedback corrections, and lower time of automatic control of posture were observed in static conditions. The four sub-groups recovered near-normal postural performance after VR in the EO and EC conditions, but still exhibited altered postural performance with optokinetic stimulation. In dynamic posturography conditions and before VR, the percentage of patients able to perform the postural tasks with EC and optokinetic stimulation was significantly lower in the two sub-groups with aVOR gain <0.20. After VR, the improvement of the postural parameters depended on the stage of rehabilitation and the degree of vestibular hypofunction. The best balance function recovery was found in the sub-group with early VR and pre-rehabilitation aVOR gain above 0.20, the worst in the sub-group with late rehabilitation and aVOR gain below 0.20. These differences were seen when the vestibular input remains the main sensory cue to control balance, that is, on unstable support without vision or altered visual motion cues. These findings extend to dynamic balance recovery the crucial roles of early rehabilitation and degree of vestibular hypofunction which we have already highlighted for vestibulo-ocular reflex recovery.