A person waits for a signal, follows a cue toward one part of a display, then chooses a response while competing information points elsewhere. It is tempting to call all three moments “attention” and leave the explanation there. Michael Posner and Steven Petersen’s 1990 paper, The Attention System of the Human Brain, asked a harder question: which functions are involved, and what evidence might connect them to systems in the brain?
This article is a critical guide to that paper. It explains the proposal that made it influential, the methods available at the time and the ways later research revised the picture. It is distinct from the Library’s entries on Michael Posner, the Attention Network Test and individual attention functions. Its job is to help a reader understand what the 1990 source established, what it proposed and what it could not yet settle.
Why the 1990 paper mattered
At the end of the 1980s, attention research had rich behavioral experiments and evidence from neurological patients. Brain imaging was beginning to let researchers ask where activity changed during carefully defined tasks. Posner and Petersen tried to join these approaches. They argued that attention should be studied as an organized system with distinguishable functions, rather than as a single supply of mental energy or a property located in whichever sensory area happened to be active.
The authors’ later review describes three commitments of the original paper. First, sources of attentional influence can be studied separately from the sensory, decision and motor processes they affect. Second, the relevant anatomy forms networks rather than one center. Third, regions in those networks can be related to functions stated in cognitive terms. Those commitments did not make the brain simple. They made the research question more precise: which manipulation of a task changes which component of selection or control, and what converging neural evidence follows?
A synthesis of this sort is not itself a trial that proves every anatomical claim. The paper assembled experimental, lesion and early imaging findings into a framework. Its lasting contribution was to make predictions and comparisons possible. Its specific assignments of functions to regions were open to correction as methods improved. That difference between a productive research program and a final atlas matters whenever a classic paper is used to explain a present-day person.
Three functions, with a historical translation
The original argument distinguished preparation or maintenance of an alert state, orienting toward relevant sensory information, and detecting or resolving a target event for focal processing. Later accounts often summarize the framework as alerting, orienting and executive attention. The names are close, but they should not be flattened into a claim that the 1990 text already contained every later task definition or every modern network boundary.
Alerting concerns readiness. A warning can prepare a rapid response, while sustained vigilance raises a different question about maintaining readiness over time. The 1990 anatomical discussion drew on brainstem arousal systems and right-hemisphere evidence. A fast response after a cue supports a statement about a tested interval; it does not reveal a person’s permanent level of alertness. Timing, fatigue, expectations and the cost of a false alarm can all change performance.
Orienting concerns shifting priority toward a place or item. Posner’s earlier cueing experiments showed that attention can move without the eyes necessarily moving at the same moment. A cue can be valid or invalid; differences in speed and accuracy provide evidence about preparing and redirecting attention under that task’s conditions. The early anatomical proposal included parietal contributions among other regions. It did not imply that one patch of cortex alone “is” orienting.
Detection and control concern what happens when a target becomes the object of focal processing and when a response must be selected amid conflict. Later writing often uses “executive attention” for this family, emphasizing conflict monitoring and control. The original paper connected parts of this work with midline frontal areas, including anterior cingulate cortex. A reader should keep the historical vocabulary visible: an early detection account and later executive-control models overlap, but they are not interchangeable sentence for sentence.
These functions interact in an ordinary task. A warning may change the effect of a spatial cue; a cue may alter how much competing information matters; a conflict may itself recruit further control. Their analytical separation is valuable because an experiment can vary one condition at a time. It is not a claim that the brain operates three sealed modules in sequence.
From behavior to an anatomical hypothesis
The paper appeared when researchers could compare reaction times, study people with localized injury and begin using positron emission tomography to investigate functional anatomy. Each method supplied a different kind of evidence. A behavioral cueing effect says a manipulation changed performance. A lesion can help test what a damaged region contributes, while also raising questions about reorganization, distributed connections and the injury’s exact extent. Imaging can reveal activity associated with a task contrast, but association is not proof that every highlighted area is necessary for the function.
Combining methods is stronger than treating any one image as a portrait of attention. If an invalid cue slows responses, the inference is not that one anatomical switch failed. The person may have allocated priority to the cued location, spent time disengaging, then reoriented. Accuracy and speed can trade off. The task may also change arousal or expectations. A good experiment compares alternative accounts, not just colored activation maps.
Posner’s 1980 discussion of orienting provided a behavioral foundation for this program. The 1990 paper carried that question into a broader systems account. What made the account useful was the mapping between operational definitions and proposed mechanisms. If “orienting” refers to a measured shift in spatial priority, researchers can ask whether changes in parietal or other networks follow that shift under multiple tasks. If the label instead means every change in what someone notices, it becomes difficult to test.
The word “separate” in the original proposal also needs care. Sources of attentional influence can be analytically distinguished from the systems they modulate. That does not mean the latter are unaffected by attention, or that one stable set of regions acts alone in every context. The later review explicitly elaborated and expanded the original framework as the imaging literature grew.
What changed over twenty years
In their 2012 retrospective, Petersen and Posner returned to the original three networks and described a much larger evidence base. They retained the value of distinguishing alerting, orienting and executive functions, while discussing additional control networks and interactions among them. In particular, later work separated aspects of sustained task control and flexible adjustment that an early three-part diagram could not fully express.
This is a normal trajectory for a useful scientific model. A framework can survive because it guides research while its anatomical details become more complex. The retrospective discusses developmental work, individual differences and candidate relations to intervention. It does not say that every new finding was already contained in 1990. Reading the two papers together shows both continuity and revision.
Modern network language can make early diagrams look cleaner than the data permit. Regions take part in multiple tasks; a task can recruit multiple networks; and the functions assigned to a network depend on a particular experimental contrast. The boundary of an “executive network” is not drawn merely by observing that a task feels difficult. Researchers need explicit conditions that distinguish conflict, working memory, error detection, response selection and sustained task maintenance.
At the same time, calling the distinctions imperfect does not make them useless. A study that separately manipulates warning cues, spatial cues and incongruent flankers can reveal patterns that one overall accuracy score would hide. The best reading treats the categories as questions about mechanisms, then asks how a given study measured each one.
The Attention Network Test came later
Fan and colleagues’ 2002 Attention Network Test combined cue and flanker conditions in a short task to estimate alerting, orienting and executive conflict effects. That instrument was developed after the 1990 review. It operationalized parts of the broader framework; it was not a table of direct readings from three brain compartments. The task’s scores are differences between specific trial conditions. A difference can be informative even when it is sensitive to practice, speed-accuracy choices or the reliability of its component measurements.
Imagine two people have the same average response time. One gains more from a warning cue; the other is slowed more by incompatible flankers. The task can make that contrast visible. But the numbers do not show that one person has a fixed alerting organ or that the other lacks self-control. Testing conditions, motivation and error rates matter. Repeated testing may change strategies, and a score formed by subtracting two noisy averages may vary between sessions.
Reliability research makes this distinction concrete. A computerized attention-battery study reviewed ANT findings and reported that alerting and orienting reaction-time difference scores often had lower reliability than executive-conflict scores in the studies it summarized. Other tasks and designs may improve measurement. The conclusion is methodological: a group effect can be clear while using one person’s difference score as a high-stakes trait label requires much stronger evidence.
For an educator or designer, the task can therefore inspire a question about readiness, spatial cues or conflict. It does not supply a ready-made diagnosis or a universal measure of a learner’s capacity. The Library’s separate ANT entry examines the task and its scoring more closely; this paper guide keeps the instrument in its historical place.
Reading a claim about brain regions
Suppose an article says that the parietal cortex “controls attention.” A careful reader asks what kind of attention was manipulated. Did participants follow a spatial cue, detect a target, keep watch over time or resolve conflicting responses? Which comparison produced the finding? Were eye movements measured or controlled? Did the study use lesion evidence, imaging, stimulation or behavior? Each answer changes what can be concluded.
A second question concerns level of explanation. “This region was more active during invalid than valid cue trials” is a narrower observation than “this region creates attention.” The latter might be a hypothesis, but it needs evidence across tasks and methods. A region’s activation can reflect reorienting, error processing, effort or another correlated event. A network account can also be revised when a new method separates processes that an old contrast joined.
A third question concerns direction of inference. If some patients with a particular injury have difficulty with orienting, the result can support a role for affected circuitry. It does not follow that everyone with slow orienting on a computer task has the same injury, nor that one brain scan can establish why a person missed a meeting instruction. Moving from controlled research to individual assessment requires validated measures and a separate clinical or educational purpose.
These questions preserve the point of the 1990 framework. Naming functions helps the experimenter avoid explaining all behavior with a single vague “attention” variable. But the names are the beginning of analysis, not an excuse to bypass measurement.
Uses in learning and digital design
The Institute’s application of this research is a design synthesis, not a claim tested in the 1990 paper. A classroom may ask students to remain ready for an instruction, orient toward a relevant source and resolve conflict when two explanations disagree. A digital service may likewise provide a warning, guide a user’s attention to a form field and require a decision under competing information. The three questions help a designer separate possible failure points.
For example, a service may put an error message far from the field it describes. The problem may be orienting: the user does not see where to act. A different service may issue urgent alerts so often that the warning loses meaning; that is closer to a question about readiness and expectation. A third may ask a person to choose between options whose descriptions conflict; the design should support comparison and correction rather than blame a person for “low executive attention.”
Those are hypotheses for testing with users. The early attention-system paper does not establish that moving an error message will improve completion rates, or that an alert policy will preserve a workforce’s attention. Usability tests, field outcomes and the people affected must decide whether a translation helps. Context matters because attention is directed toward purposes that may be more important to the person than the experimenter’s target.
That final point is easy to miss in laboratory diagrams. A participant follows a task instruction for a brief session. A learner or service user may have multiple legitimate goals, interruptions, health constraints and unequal access to time. An ethically sound design should not turn a functional network description into a judgment about character or worth.
What the evidence does not show
The 1990 paper did not prove a one-to-one map between three psychological words and three fixed locations. It did not measure a universal attention capacity, establish a ranking of people or show how to train every network. It proposed a framework from the evidence then available. Later research strengthened some parts, subdivided others and exposed interactions that a simple diagram cannot show.
Neither the 1990 synthesis nor the later ANT justifies diagnosing an individual from a few response times. A score depends on its task contrast, and its usefulness for a consequential decision depends on reliability, validity and the population in which it is used. A striking brain image or the phrase “executive network” cannot replace that work.
Nor does the framework say that attention is wholly voluntary. A warning can alter readiness without making every response correct. A valid cue can help orienting without compelling a person to look. Conflict may require control even when the person knows the right rule. The paper gives a vocabulary for investigating these differences, not a promise that a sufficiently motivated person can make them disappear.
Return to the task display
The person now sees a second trial. A warning comes earlier, the spatial cue is misleading and the competing response is stronger. An overall delay tells us little until we compare the conditions. Which part of the delay reflects readiness, which reflects reorienting and which reflects conflict? Posner and Petersen’s lasting invitation was to make those questions testable, then revise the map when better evidence arrives.
The value of the 1990 paper is therefore neither nostalgia nor an anatomical slogan. It turned a broad word into a disciplined research program. Readers can honor that achievement by preserving the paper’s historical limits and using the later evidence to ask more exact questions about the next person, task and setting.
Research references
- Posner, M. I., & Petersen, S. E. (1990). The attention system of the human brain. Annual Review of Neuroscience, 13, 25–42. View source
- Petersen, S. E., & Posner, M. I. (2012). The attention system of the human brain: 20 years after. Annual Review of Neuroscience, 35, 73–89. View source
- Posner, M. I. (1980). Orienting of attention. Quarterly Journal of Experimental Psychology, 32(1), 3–25. View source
- Fan, J., McCandliss, B. D., Sommer, T., Raz, A., & Posner, M. I. (2002). Testing the efficiency and independence of attentional networks. Journal of Cognitive Neuroscience, 14(3), 340–347. View source
- Ishigami, Y., & Klein, R. M. (2011). Repeated measurement of the components of attention of older adults using the two versions of the Attention Network Test: Stability, isolability, robustness, and reliability. Frontiers in Aging Neuroscience, 3, 17. View source
- Jones, S. A. H., Butler, B. C., Kintzel, F., Johnson, A., Klein, R. M., & Eskes, G. A. (2016). Measuring the performance of attention networks with the Dalhousie Computerized Attention Battery: Methodology and reliability in healthy adults. Frontiers in Psychology, 7, 823. View source