Considered is the optimal processing of multisensory information to maintain body orientation in the human. It is supposed that the orientation is performed by a system of angular stabilization with only small deviations from the equilibrium believed to be coincided with the local vertical. The linear transformation is determined which minimizes the square loss function for the stabilization error. This transformation comprises two successive steps. One of them is the averaging procedure on the the sensory inputs and the other is the optimal Kalman filtering in time domain. It is shown that the presence of two interacting channels for angular position and angular velocity in the structure of the optimal filter ensures sufficient accuracy of stabilization in both channels for the stabilization system as a whole in spite of the substantial unreliability in transmission in the channel of the angular position.