In two experiments, participants decided on each of several trials whether or notto take a risk. If they chose to take the risk, they had a relatively highprobability (from 75% to 95%) of winning a small number of points and arelatively low probability (5% to 25%) of losing a large number of points. Theloss amounts varied so that the expected value of taking the risk was positiveon some trials, zero on others, and negative on the rest. The main independentvariable was whether the probability of losing was communicated using numericalpercentages or icon arrays. Both experiments included random icon arrays, inwhich the icons representing losses were randomly distributed throughout thearray. Experiment 2 also included grouped icon arrays, in which the iconsrepresenting losses were grouped at the bottom of the array. Neither type oficon array led to better performance in the task. However, the random iconarrays led to less risk taking than the numerical percentages or the groupedicon arrays, especially at the higher loss probabilities. In a third experiment,participants made direct judgments of the percentages and probabilitiesrepresented by the icon arrays from Experiment 2. The results supported the ideathat random arrays lead to less risk taking because they are perceived torepresent greater loss probabilities. These results have several implicationsfor the study of icon arrays and their use in risk communication.