Innovations in software for the analysis of eye movements have not kept pace with the develop ment of hardware for collecting samples of eye position (Scinto & Barnette, 1986). Eye fixation and duration have been the primary measures for gleaning knowledge of subjects' performance while the subjects are engaged in cognitive visual tasks. The program Cluster was developed as a means of investigating the dynamics of target examination characteristics that do not lend themselves to traditional methods of eye-movement analysis. This tool has proved to be a valu able means of assessing visual activity at a micro level, in comparison with the gross measures of distribution of visual attention in various areas of the visual field. In this report, we describe the history surrounding the development of Cluster as an analytical tool, the source of input re quired for its execution, the mechanics of the execution as an interactive process, the replicabil ity of raters' judgments, the program's products of visual displays and data file output, and the potential application of such a tool for analyzing visual activity. Scientific research does not always adhere to ex perimental protocol in a simple, straightforward manner. In the statistical analysis of the data obtained from even a well-designed research study, unexpected aspects of human behavior and performance are often exposed. In the present paper, we describe a computer program de signed and written to handle just such a situation, which arose during a research endeavor (Birkmire, Karsh, Bar nette, Pillalamarri, & Breitenbach, 1991) of the U.S. Army Human Engineering Laboratory (HEL) to aid designers of military tracked vehicles at the U.S. Army Tank-Automotive Command (TACOM). Specifically, these designers were interested in determining what ve hicle design features (e.g., gun barrels, turrets, etc.) were more or less critical in the identification of a vehicle viewed through an electronic imaging device, such as a forward-looking infrared (FUR) sensor. HEL's visual performance team developed a research program to ascertain whether eye movements could be useful for the attempt to determine what vehicle features might be critical for target discrimination. We designed an experiment in which we used a method developed and reported by Gerhart, Graziano, and Carter (1983), whereby a stored image of a vehicle silhouette could be combined with a background of digitally generated Gauss ian noise. The result is a quantifiable simulation of a non specific sensor system image. The subject's task was to visually examine a series of these images to identify one of eight different targets embedded in varying levels of target signal-to-background noise ratios (SNR).