We introduce visual graphs as an intermediate repren between concrete visual syntax and abstract graph syntax. In a visual graph some nodes are shown as geometric figures, and some edges are represented by geometric relationships between these figures. By carefully designing visual graphs and corresponding mappings to syntax graphs, semantics definitions can, at least partially, employ a visual notation while still based on syntax. Visual semantics thus offers the "best of lds" by integrating abstract syntax and visual notation. These concepts can also be used to give visual semantics for traditional textual formalisms. As an example we provide a visual definition of Turing machines.