Top-down vs bottom-up processing: the difference, and why perception needs both

Bottom-up processing is perception that the stimulus drives: your senses detect basic features, and your brain builds them into a percept. Top-down processing is perception that knowledge and expectation drive: what you already know, the context and your goals shape how you interpret the input. A loud crash that pulls your attention away from a conversation is bottom-up. Seeing the same ambiguous shape as a B in a row of letters and as 13 in a row of numbers is top-down. Normal perception uses both.

The two terms are easy to confuse. Both describe the same act, perceiving, and both names give a direction, not a thing. This article gives the definitions from an open textbook and from the AP Psychology course. It then covers the classic studies behind each direction, the part of the debate that is still open, and a method to sort exam scenarios.

• Bottom-up processing starts with the input. Top-down processing starts with what the perceiver brings: knowledge, expectations, context and goals.
• The AP Psychology course framework treats the difference as one of degree: perception "primarily relies" on one source or the other.
• In a 1970 study by the psychologist R. M. Warren, listeners heard a speech sound that a cough had replaced. When silence replaced the sound, they noticed the gap.
• Brain research describes feedback pathways that carry expectation into the visual cortex. So in normal perception, the two directions work together.
• Researchers still argue how far expectations change what we see, as opposed to what we judge or report.

The difference in one table

The definitions come from the open textbook Psychology 2e by OpenStax. It defines bottom-up processing as "sensory information from a stimulus in the environment driving a process", and top-down processing as "knowledge and expectancy driving a process". The same section adds that "perception involves both bottom-up and top-down processing."

The College Board uses the same pair in Topic 2.1 of its AP Psychology course and exam description. Perception, it says, depends on whether a person "primarily relies on external sensory information (bottom-up processing) or internal prior expectations (top-down processing)." The word "primarily" matters. An exam answer names the process that plays the main part. It does not claim that the other one is absent.

Bottom-up processing: the stimulus drives it

In bottom-up processing, information flows from the input upward. OpenStax describes it as the case where "we sense basic features of stimuli and then integrate them". The perceiver does not need to know what the object is. The features of the stimulus carry the process.

Attention gives a clear example. Suppose that you are in a noisy restaurant and you concentrate on one conversation. Then a loud crash breaks through, and your attention goes to the sound. OpenStax calls this attentional capture and says that "it would be bottom-up", because the sound in the environment caused it. You did not decide to listen for a crash.

Bottom-up and top-down processing are both parts of perception. OpenStax defines perception as "the way sensory information is organized, interpreted, and consciously experienced". Sensation is the detection itself, by the sensory receptors. So the two terms do not divide sensation from perception. They name two sources of information inside perception.

Top-down processing: knowledge and expectation drive it

In top-down processing, what you already know shapes what you perceive. OpenStax describes it as the case where "previous experience and expectations are first used to recognize stimuli".

The textbook example is a shape that can be a letter or a number. In a row of letters, people see it as a B. In a row of numbers, the same lines look like 13. The input does not change. Only the context changes, and the context decides what you see.

The psychologist R. L. Gregory gave the classic account of this view. In a 1980 paper, he wrote that "perceptions may be compared with hypotheses in science". In his account, the brain processes neural signals with stored knowledge and assumptions, and the result is a hypothesis about what is out there.

OpenStax uses the same idea in its section on Gestalt principles. It says: "Our perceptions are based on perceptual hypotheses: educated guesses that we make while interpreting sensory information." Personality, experience and expectations inform these guesses, and the guesses generate a perceptual set. The AP course lists "schemas and perceptual sets" as "internal factors that filter perceptions of the world".

Top-down processing also guides search. If you lose your keys, you look in the places where you usually leave them, not on the ceiling fan. OpenStax gives this as a top-down example. In the context of attention, it describes top-down processes as "generally goal directed, slow, deliberate, effortful, and under your control".

Expectations also affect how we judge other people, for example in the Pygmalion effect and other expectancy effects. That is a related but separate topic, covered in our article on perceptual errors in judging people.

Two classic studies that show both at work

Warren (1970): a cough that hides a speech sound

Warren took a recorded sentence and replaced one speech sound with a different sound, such as a cough or a tone. According to his 1970 report in Science, listeners "believe they hear the missing sound". They also could not place the cough correctly. It seemed to occur during another part of the sentence. The title of his report calls this the "perceptual restoration of missing speech sounds".

The study had one more condition. When silence replaced the speech sound, listeners located the gap correctly and noticed that the sound was missing. So the sentence filled in the sound only when the input allowed it. A cough could hide a speech sound, but silence could not. Top-down processing worked on the ambiguous input. When the input was clear, bottom-up processing decided.

Palmer (1975): a scene, then a brief look at an object

Stephen Palmer first showed people a picture of a scene. Then he showed a drawing of an object for a very short time. The object fitted the scene, did not fit it, or came after no scene at all. According to Palmer's 1975 report, people named the object correctly most often when it fitted the scene, and least often when it did not. The context changed which object people reported.

The confidence of each answer depended on how similar the drawing was to the object that the person named. It did not depend strongly on the context. So the context changed the answer, but the visual features still controlled how sure people were.

Why perception needs both

Most of the time, the two directions work together, not against each other. Brain research supports this view. In a 2013 review, Charles Gilbert and Wu Li describe feedback pathways between areas of the visual cortex. These pathways carry information about attention, expectation, the current task, working memory and motor commands. As a result, the same neurons can work in different ways for different tasks.

One influential account goes further. The philosopher Andy Clark summarised the view that brains are prediction machines. In this view, the brain constantly tries "to match incoming sensory inputs with top-down expectations or predictions", in "a bidirectional cascade of cortical processing". Clark called the approach the best clue yet to a unified science of mind, and he also listed its open problems. It is a framework under debate, not a settled fact.

Where the line is still argued

According to Chaz Firestone and Brian Scholl, a "growing consensus" holds that top-down effects on seeing are everywhere. In 2016, they reviewed what they called "hundreds of studies" that claimed beliefs, desires, emotions or language change what people see. In their article in Behavioral and Brain Sciences, they argued that none of these studies gives compelling evidence for a true top-down effect on perception itself.

They named six pitfalls that, in their view, explain the reported effects. One is the difference between perception and judgement. You can see the colour or size of a shoe, but you can only judge that the shoe is expensive. So when an experiment changes what people report, the change can be in their judgement, not in what they see. Errors of judgement are a large field of their own, covered in our article on cognitive biases. The other five are an overly confirmatory research strategy, demand and response bias, low-level differences between the stimuli, peripheral attention effects, and memory and recognition.

One of their own studies shows the low-level pitfall. An earlier report claimed that people see a Black face as darker than a White face with the same average brightness. In a 2015 study, Firestone and Scholl showed heavily blurred versions of the faces. Observers still judged the Black face as darker, even when they could not see the race of the faces. The difference came from features of the images themselves, which is a bottom-up cause.

Their view is one side of a live debate. The journal published their article with replies from other researchers, and the authors answered them. Gregory saw a similar problem in 1980. He wrote that it is "surprisingly difficult" to tell whether physiological or cognitive factors cause even the classical distortion illusions. For students, the textbook distinction stays the same. The open question is how deep top-down effects go: into seeing itself, or into what people judge and report.

Classifying exam scenarios

Two questions sort most exam scenarios:

1. 1.What starts the process: the stimulus, or what the person knows, expects or wants?
2. 2.Would a person without that knowledge or expectation perceive the same thing?

If the stimulus starts the process and anyone would perceive the same thing, the answer is bottom-up. If knowledge, context or expectation changes the result, the answer is top-down.

Some questions describe both processes. Then look for the word that the question stresses. "Because she expected", "in this context" and "from experience" point to top-down processing. "Before he recognised it" and "the brightness of the light" point to bottom-up processing.

How to keep the two terms apart

A simple memory aid: the bottom is the input, and the top is the brain. Bottom-up processing starts at the bottom, with the raw signals from the senses, and builds up to a percept. Top-down processing starts at the top, with what the brain already holds, and works down to the input. A short version is "senses first" for bottom-up and "knowledge first" for top-down.

The same words also appear in discussions of therapy, ADHD and autism. Those uses are outside this article, which covers the meaning in perception research and psychology courses.

The vocabulary of this topic is a small set of related terms: bottom-up processing, top-down processing, perceptual set, perceptual hypothesis, schema and attentional capture. In Psychology World, you look up a term and save it to one of your own lists. Then you drill the list as flashcards, by typing the answer, or as multiple-choice questions once the list has four or more terms. The four steps of the learning loop and what each study mode does are on the home page. For study methods with strong evidence behind them, read our guide to memorising psychology terminology.