Fingertip contact with another person reduces sway(1). However, this represents a complex control loop, with both persons relying on a moving reference point. Here we attempt to understand this control loop with a combination of experiment and modelling. We also determine whether vision of another person drives sway, and if this interacts with touch. We studied 8 pairs of subjects standing next to each other on separate forceplates. We included 3 conditions of touch (no contact(NC), light touch(LT), shoulder contact (SC)) and 3 visual conditions (both eyes closed, both eyes open, asymmetric). Centre of pressure velocity (COPv) was the measure of sway. Magnitude and timing of sway coupling was quantified by the COPv-COPv cross-correlation function (XCORR). Data was compared with a Simulink model(2) consisting of a two inverted pendulums and PID controllers coupled together. Physical contact (LT & SC) reduced sway, with a greater effect for shoulder contact (Fig 1A; F=193,21;p<0.01). Vision also reduced sway but less so with increasing tactile contact (p<0.01). When only light touch was available, XCORR’s exhibited twin peaks at ~500 and -500ms (Fig 1B). Vision alone produced strikingly similar XCORR’s. However, the effect of touch and vision did not summate with both available. During asymmetric vision, XCORR’s showed only 1 peak at positive lag. In contrast to light touch, shoulder contact produced XCORR’s with a single peak at ~0ms lag irrespective of visual condition. We modelled the interaction in simulink as follows. Firstly, we coupled both sensory feedback loops to each other, such that each (model) person’s feedback error was summed with 20% of their partner’s. This recreated the XCORR shape and timing observed during both visual and light touch conditions (Fig 1C). Secondly, we coupled the torque output, such that 20% of one person’s ankle torque summed with the other (while maintaining the sensory coupling). This recreated the XCORR seen during shoulder contact. Firstly we conclude that tactile sway coupling is a continuous bidirectional sensorimotor process with a delay in each direction of ~500ms. It is unnecessary to invoke ‘switching’ mechanisms to explain the interaction, whereby a sway leader and follower periodically swap roles; a continuous control model was sufficient to recreate the XCORR characteristics. Unlike light touch, effects of shoulder contact were explained by mechanical linkage. Secondly, we have shown that vision of another person produces sway coupling near identical to touch. This is despite only peripheral vision of one’s partner, and abundant veridical feedback from the laboratory. The observation that the effects of touch and vision did not summate suggest sensory redundancy. Paradoxically, the visual coupling demonstrates that sometimes the partner with eyes closed lead the sighted partner.