Three experiments tested the hypothesis that graphs convey information effectively because they can display global trends as geometric patterns that visual systems encode easily. A novel type of graph was invented in which angles/lengths of line segments joined end-to-end represented variables of rainfall and temperature of a set of months. It was expected that questions about single values of a variable would be easier to answer when the variable was encoded as a segment length whereas questions about global trends of a variable would be easier to answer when the variable was encoded as a segment angle. Subjects' (N=14) response times when answering questions pertaining to such graphs demonstrated the interaction hypothesized. This was true both when subjects construed stimuli as meaningless visual patterns and when they construed the same stimuli as graphs. Similar results were obtained regardless of whether subjects were explicitly instructed about how trends of the angle variable translated into geometric shapes. It is concluded that graph formats, and types of information conveyed by graphs, are not uniformly easy or difficult, but that a given type of information is conveyed efficiently in a graph format to the extent that it corresponds to a naturally perceivable visual pattern. (Author/JN)