It is argued that vestibular internuclear commissural pathways are functionally important in the vestibuloocular reflex (VOR), particularly since they appear to be modulated during nystagmus. A bilateral approach to VOR modeling is essential to an effective study of the effects of commissural connections on response dynamics. A bilateral model of the VOR central pathways is proposed, with three main postulates: neural filters (NF) on each side of the brain stem, each linked to tonic cells in the ipsilateral vestibular nuclei in negative feedback loops; strong coupling between these bilateral loops by reciprocal commissural connections that significantly affect response dynamics; and modulation of this coupling by inhibitory burst neurons during fast phases. Mathematical analysis of this model shows that the NF need not be good integrators. During slow-phase operation, commissural pathways provide a positive-feedback effect that improves the effective integration function of the bilateral system beyond that of the NF in each side. Analysis suggests that the time constant of the NF might even be as small as that of the eye plant (approximately 0.24 s), so that the NF might be considered to be internal models of the eye plant rather than pseudointegrators. In the model, modulation of commissural gains by burst cells is shown to be sufficient to cause the system to switch between a compensatory position-tracking mode (slow phases) and an anticompensatory velocity-tracking mode (fast phases) during nystagmus. The model simulates a number of behavioral and neurophysiological findings, such as a) tonic vestibular nuclei (VN) cells have sensitivities and decay times larger than primary vestibular fibers, and their response polarity may reverse after section of superficial commissural fibers; b) effective VOR integration deteriorates after cerebellectomy or commissurectomy; c) peak fast-phase eye velocity is modulated by the vestibular signal as well as by fast-phase amplitude. The model accounts for the modulation of central VN responses during nystagmus and, as a result, simulations strongly imply that envelopes of slow-phase eye velocity or smoothed central firing rates will depend on fast-phase strategy and, hence, may not always yield accurate estimates of VOR dynamics. Similarly, the model predicts that "apparent" disassociation between central and ocular responses may occur because of interactions during nystagmus, despite appropriate behavior within slow-phase segments (since VN responses are not simple estimates of eye velocity).(ABSTRACT TRUNCATED AT 400 WORDS)