Icons are a very important component of graphical user interfaces. However,
icon design is still predominantly artistic in nature and as a result icon
selection is generally based on usability evaluations after a set of alter
native icons are developed. This process tends to be time-consuming and cos
tly. In this research, we address the issues of what should be depicted in
an icon, given the function it should represent, and how training affects t
he performance of novice users when using an iconic interface. A set of 36
concrete icons (12 functions) were selected and tested with a total of 30 p
articipants. The experimental results indicate that complete representation
s are generally superior for both untrained and trained participants. Resul
ts also show that trained participants had shorter response times when comp
ared to untrained participants. Further analysis suggests that ambiguity, u
niqueness and dominance are three important aspects to consider when design
ing and developing icons. Applications of this research include the design
of appropriate icons for graphical user interfaces prior to usability testi
ng and the importance of a short training period to illustrate the composit
ion of an icon in an effort to improve the mental model associated with eac
h design. (C) 2001 Academic Press.