Research on graphic symbol learning suggests that symbols with a greater visual resemblance to their referents (greater iconicity) are more easily learned. The iconicity hypothesis has not yet been explored within the intervention protocol of the Picture Exchange Communication System (PECS). Within the PECS protocol, participants do not point to a symbol but exchange the symbol for an object. The purpose of this study was to examine whether children learn to request more readily with PECS when the symbols involved are highly iconic versus symbols that are low in iconicity. An adapted alternating treatments design combined with a multiple baseline design across subjects was used to evaluate the effectiveness and efficiency of symbol learning under two conditions: high iconicity and low iconicity. Four students with autism or pervasive developmental disorders between the ages of six and nine years participated. Results indicated that students learned to request desired objects under both conditions, lending further support for the effectiveness of PECS. There was little to no difference, however, in the effectiveness and efficiency of requesting between the two conditions during Phases I and II of PECS training. Thus learners do not benefit from symbols that bear more resemblance with their referents during the first two phases of PECS instruction. (Contains 1 table and 2 figures.)