The biased competition model, developed by Robert Desimone and John Duncan in the 1990s, provides a neural account of selective attention. Rather than treating attention as a spotlight or filter, it proposes that attention emerges from competitive interactions among neural populations representing different stimuli. When multiple objects fall within a neuron's receptive field, their representations compete — and top-down attentional signals bias this competition in favor of task-relevant objects.
Core Principles
The model rests on several key principles. First, when multiple stimuli are present in a neuron's receptive field, they mutually suppress each other's neural responses — a phenomenon called competitive interaction. Second, this competition can be biased by top-down signals from frontal and parietal areas that represent current behavioral goals. Third, the "winning" stimulus representation is enhanced while competing representations are suppressed. Fourth, these competitive interactions operate throughout the visual hierarchy, with the effects of biasing propagated from higher areas back to lower areas.
Response to B alone: R(B)
Response to A + B (unattended): R(A+B) < R(A) — suppression
Response to A + B (attend A): R(A+B|attend A) ≈ R(A) — bias resolves competition
Neural Evidence
Single-unit recording studies in monkeys, particularly by Desimone and colleagues in area V4 and inferotemporal cortex, provided direct evidence for the model. When two stimuli (one effective and one ineffective for a given neuron) were presented within the receptive field, the response was intermediate — consistent with competition. When attention was directed to the effective stimulus, the response increased toward the level evoked by the effective stimulus alone — consistent with biased competition resolving in favor of the attended stimulus.
John Duncan extended the biased competition framework into the integrated competition hypothesis, proposing that attention operates as object-based selection across multiple brain systems simultaneously. When attention selects an object, all its properties (color, shape, location, identity) are enhanced across the relevant cortical areas. This explains why attending to one feature of an object (its color) also enhances processing of its other features (its shape), supporting object-based theories of attention.
Relationship to Other Theories
The biased competition model provides a neural mechanism that can accommodate many insights from information-processing theories. Early selection (Broadbent) corresponds to biasing at lower visual areas; late selection corresponds to biasing at higher areas. Feature-based attention corresponds to bias signals targeting feature-selective neurons; spatial attention corresponds to bias targeting location-specific neurons. The model thus serves as a unifying neural framework for diverse attention phenomena.
Computational Implementations
The biased competition model has been implemented in computational neural networks, most notably by Heinke and Humphreys (SAIM model) and by Reynolds and Heeger (normalization model of attention). These models demonstrate that competitive interactions combined with top-down biasing can account for a wide range of attentional effects, including changes in neural response gain, contrast sensitivity, and receptive field properties.