The paper presents a revised model of visual search, Guided Search 2.0 (GS2), which builds on earlier work by Neisser, Treisman, Julesz, and others. The model distinguishes between a preattentive, massively parallel stage that processes basic visual features (color, motion, depth cues) across the visual field and a subsequent limited-capacity stage that handles complex tasks (e.g., face recognition, reading) over a limited portion of the visual field. The limited-capacity stage is under attentional control, and its deployment is guided by the output of the earlier parallel processes. GS2 is a revision of the original model, incorporating new data and refining aspects of the model. The paper is organized into four parts. Part 1 describes the model and its simulation. Part 2 reviews preattentive processing of basic features and shows how the simulation reproduces these results. Part 3 reviews attentional deployment in conjunction and serial searches and shows how the simulation handles these conditions. Part 4 discusses shortcomings of the model and unresolved issues. The model assumes that input and initial processing of stimuli are carried out in parallel across the entire visual field. At some point, independent parallel representations are generated for a limited set of basic visual features, called feature maps. These maps can be structured in different ways, but GS2 follows the second alternative of one map per feature type. Bottom-up activation is a measure of how unusual an item is in its context. It is calculated separately for each feature and depends on the differences between an item and its neighbors. Top-down activation is user-driven and is used to specify the desired value of the target for a given feature. The correct channel is selected based on the difference between the target and distractors. The activation map combines information from bottom-up and top-down processes. Attention is deployed in order of decreasing activation. The model also includes a termination rule for unsuccessful searches, based on a staircase adjustment procedure. The model assumes that the activation of the target is a signal plus noise, and that the efficiency of search depends on the strength of the attention-attracting signal of the target relative to the noise of the distractors. The model is implemented as a computer simulation, which can produce data that closely resemble human data for a wide variety of standard search tasks.The paper presents a revised model of visual search, Guided Search 2.0 (GS2), which builds on earlier work by Neisser, Treisman, Julesz, and others. The model distinguishes between a preattentive, massively parallel stage that processes basic visual features (color, motion, depth cues) across the visual field and a subsequent limited-capacity stage that handles complex tasks (e.g., face recognition, reading) over a limited portion of the visual field. The limited-capacity stage is under attentional control, and its deployment is guided by the output of the earlier parallel processes. GS2 is a revision of the original model, incorporating new data and refining aspects of the model. The paper is organized into four parts. Part 1 describes the model and its simulation. Part 2 reviews preattentive processing of basic features and shows how the simulation reproduces these results. Part 3 reviews attentional deployment in conjunction and serial searches and shows how the simulation handles these conditions. Part 4 discusses shortcomings of the model and unresolved issues. The model assumes that input and initial processing of stimuli are carried out in parallel across the entire visual field. At some point, independent parallel representations are generated for a limited set of basic visual features, called feature maps. These maps can be structured in different ways, but GS2 follows the second alternative of one map per feature type. Bottom-up activation is a measure of how unusual an item is in its context. It is calculated separately for each feature and depends on the differences between an item and its neighbors. Top-down activation is user-driven and is used to specify the desired value of the target for a given feature. The correct channel is selected based on the difference between the target and distractors. The activation map combines information from bottom-up and top-down processes. Attention is deployed in order of decreasing activation. The model also includes a termination rule for unsuccessful searches, based on a staircase adjustment procedure. The model assumes that the activation of the target is a signal plus noise, and that the efficiency of search depends on the strength of the attention-attracting signal of the target relative to the noise of the distractors. The model is implemented as a computer simulation, which can produce data that closely resemble human data for a wide variety of standard search tasks.