Since the time of Euler, diagrammatic tools have been used for logical reasoning. More recently, visual languages have been developed in computer science, for programming, specification, database querying, and so on. I shall examine the ``efficacy'' of such languages, applying formal semantics and complexity theory to determine their expressive power and computational properties. I show that inefficacy (in the sense of incorrectness) arises for many visual languages -- for example; Euler's Circles, linear diagrams for syllogisms (Englebretsen 1992), and a new graphical database query tool (developed by Xerox). The results point to a set of formal requirements on the design of effective visual languages.