Symbols whose sizes code a variable are often superimposed on a map to show how the value of the variable changes geographically. In many cases the areas of the symbols are proportional to the values of the variable. But previous psychological experiments indicate that people's judgments of area typically are not proportional to area, but rather to area raised to a power less than one. In many of these studies, however, the symbols are presented one at a time. Here we describe an experiment that tests whether this experimental laboratory result extends to statistical maps. We had 24 viewers judge sets of circles presented simultaneously, either with or without maplike grid ticks, labels, scale, and border. With these multiple-circle displays, judgments were related to actual area by power functions with an average exponent close to one, implying that it is appropriate to code symbols by area on statistical maps.