Many techniques for visualization and interaction that potentially increase user performance have been studied in the growing field of virtual reality. However, the effects of virtual-arm representations on users' performance and perception in selection tasks have not been studied before. This paper presents the results of a user study of three different representations of the virtual arm: "hand only," "hand+forearm," and "whole arm" which includes the upper arm. In addition to the representations' effects on performance and perception in selection tasks, we investigate how the users' performance changes depending on whether collisions with objects are allowed or not. The relationship between the virtual-arm representations and the senses of agency and ownership are also explored. Overall, we found that the "whole arm" condition performed worst.