A method to represent first-order predicate logic (Horn clause logic) by a data-flow network is presented. Like a data-flow computer for a von Neumann program, the proposed network explicitly represents the logical structure of a declarative program by unlabeled edges and operation nodes. In the deduction, the network first propagates symbolic tokens to create an expanded AND/OR network by the backward deduction, and then executes unification by a newly developed method to solve simultaneous equations buried in the network. The paper argues the soundness and completeness of the network in a conventional way, then explains how a kind of ambiguous solution is obtained by the newly developed method. Numerical experiments are also conducted with some data- flow networks, and the method’s convergence ability and scaling property to larger problems are investigated.