Three methods of labeling graphs were compared: direct labeling of the functions; a key inserted on the graph field below the functions; and a key inserted below the figure in the position of the figure caption. In both a separate-groups comparison and a subsequent within-subjects comparison, direct labeling reliably gave the quickest readings ( p <;0.01) without loss of accuracy. Reading the labels directly on the functions appears to involve fewer steps and to depend less on short-term memory.