Treisman and Gelade (1980)

Features are encoded automatically and in parallel across visual scene, but the accurate combination of features into objects requires focal, serial attention. Focal attention is the "glue" that integrates separable features into unitary objects. Top-down processing can also combine features, but not always accurately, resulting in illusory conjunctions.

Logic: If FIT makes usable predictions in all paradigms, even if not perfect, it lends credence to the theory. 1) visual search: classic feature/conjunction based search to show parallel vs. serial (affect of adding number distractors)2) texture segregation: ts and figure ground grouping should be determined by spatial discontinuities of features, not conjunctions3) illusory conjunctions: should happen when processing time is limited or attention is elsewhere4) identity and location: independent for features, loc must proceed ID for conjunctions5) interference from unattended stimuli: unattended stim should be registered at feature level only

Features: detectable by parallel search, give rise to illusory conjunctions, detected without locating, mediate testure segregation, have behavioral effects when unattended
Conjunctions: require serial search, no affect on performance unless attended, yield correlated performance on loc/ID tasks, ineffective for texture/group segmentation

Replicate classic results with addition that feature search requires monitoring two dimensions.
Results were replicated: As display size increased, feature search RTs remained constant, but conjunction search RTs grew linearly. Target present RTs had half the slope of target absent RTs, indicating serial, self-terminating scanning.

Sought to determine relationship between discriminability of features defining conjunction and speed of search.
Results: search slopes steeper for more difficult conjunctions

Determined if difficulty of conjunction search was because they share a feature with all nontargets (unlike features, which only share feature with half of NT)
Found this is not the reason conjunction search is more difficult

• search for letters serial (linear search slope) only if distractors can be recombined into target, simply “similar” letters (searching for R in P/B) is, according to them, parallel, even though search slope isn’t flat (decelerating linearity)

• texture segmentation is parallel only when boundary is defined by a single feature, conjunction boundaries harder to spot

• -without focal attn, might identify feature but not localize it
• -found that this is the case for feature targets