We present our approach to optical dynamic interconnections for neural networks. In this approach we fabricate waveguide structures in photorefractive media by simply focusing and scanning a laser beam. Resultant waveguides with variable index profiles are used for the optical dynamic interconnections. We show the experimental results of waveguide fabrication in a volume of a photorefractive (PR) medium and present the optical dynamic characteristics of the PR waveguide. We propagated the green Ar ion laser light (515 nm) in the photorefractive waveguide repeatedly and observed the variation of the near-field pattern at the end face of a 2 mm-thick lithium niobate (LiNbO3; LN) crystal. The average optical intensity of the guided beam decayed exponentially with respect to the total exposure over the waveguide. It is supposed that the guided beam redistributed or destroyed the charge density field formed by the initial exposure. A numerical analysis based on the beam propagation method and a simple PR waveguide model of this process showed reasonable agreement with the experimental results. This variation of the guided optical power can be used to implement dynamic interconnections for neural networks.