Two experiments evaluated the properties of polygon displays and bar graphs as fault indicator for systems with many parameters. A modified visual search paradigm was used to test the effectiveness of different display configurations to be check-read for the presence of abnormal ('target') readings. Whether the task was to detect the occurrence of a single abnormal (off-limits) parameter or count the number of abnormal parameters, both displays yielded response times and error rates that were independent of the total number (from 4 to 16) of displayed parameters. When the task was fault detection, the subjects performed equally well with both types of displays. When the task was counting the number of abnormalities, performance with the bar graph was independent of the number of abnormalities but performance with the polygon display was poorer overall and deteriorated with larger numbers of abnormalities. The results contradict either the proximity-compatibility hypothesis of Wickens or the traditional classification of polygons and bar graphs as typical integral and separable displays, respectively. The results are best characterized in terms of the similarity relations between and among 'target' and 'nontarget' parameters.