linearly with the number of distractors. It is as if the subjects who are placed in
thesecircumstancesareforcedtofocusattentioninturnoneachiteminthe
display in order to determine how the item’s properties or parts are conjoined.
In a positive trial (a trial in which a target is present) the search ends when the
target is found; on the average, therefore, it ends after half of the distractors
have been examined. In a negative trial (in which no target is present) all the
distractors have to be checked. As distractors are added to the displays, the
search time in positive trials therefore increases at half the rate of the search
time in negative trials.
The difference between a search for simple features and a search for con-
junctions of features could have implications in industrial settings. Quality-
control inspectors might, for example, take more time to check manufactured
items if the possible errors in manufacture are characterized by faulty combi-
nations of properties than they do if the errors always result in a salient change
in a single property. Similarly, each of the symbols representing, say, the des-
tinations for baggage handled at airline terminals should be characterized by a
unique combination of properties.
In a further series of experiments on visual-search tasks, we explored the ef-
fect of exchanging the target and the distractors. That is, we required subjects
to find a target distinguished by the fact that itlacksa feature present in all the
distractors. For example, we employed displays consisting ofOsandQs, so
that the difference between the target and the distractors is that one is simply a
circle whereas the other is a circle intersected by a line segment (figure 16.4).
We found a remarkable difference in the search time depending on whether
the target was theQand had the line or was theOand lacked the line. When
the target had the line, the search time was independent of the number of dis-
tractors. Evidently, the target popped out of the display. When the target
lacked the line, the search time increased linearly with the number of dis-
tractors. Evidently, the items in the display were being subjected to a serial
search.
The result goes against one’s intuitions. After all, each case involves the same
discrimination between the same two stimuli:OsandQs. The result is consis-
tent, however, with the idea that a pooled neural signal early in visual pro-
cessing conveys the presence but not the absence of a distinctive feature. In
other words, early vision extracts simple properties, and each type of property
triggers activity in populations of specialized detectors. A target with a unique
property is detected in the midst of distractor items simply by a check on
whether the relevant detectors are active. Conversely, a target lacking a prop-
erty that is present in the distractors arouses only slightly less activity than a
display consisting exclusively of distractors. We propose, therefore, that early
vision sets up a number of what might be calledfeaturemaps.Theyarenot
necessarily to be equated with the specialized visual areas that are mapped by
physiologists, although the correspondence is suggestive.
We have exploited visual-search tasks to test a wide range of candidate fea-
tures we thought might pop out of displays and so reveal themselves as prim-
itives: basic elements in the language of early vision. The candidates fell into a
number of categories: quantitative properties such as length or number; prop-
erties of single lines such as orientation or curvature; properties of line arrange-
404 Anne Treisman