Top-Down Visual Perception: How You Can Tap Memory To Direct Eye Movement

Why does some visual information stick with us for years, while other information is gone in an instant? How does the visual data we take in get stored for later use? How does memory suggest where to look next?

Last week I started talking about visual perception and mentioned that a 2-way process was taking place. We addressed one of the directions last week when we talked about bottom-up processing.

This week I want to consider the other direction, top-down processing. I want to talk about memory and how visual information finds its way in. I also want to consider how information held in memory works to influence where we look next and what information we process.


The Role of Memory in Visual Perception

If you remember last week’s post, when we take in our visual world our eyes are quickly darting about and then fixating momentarily to take in what little information can be clearly seen by our fovea. We process the data and then move our eyes and fixate on a new area.

We move our gaze, take in what information we can and move again. We can only hold so much information at a given time. If you’d like to hold information for the long term, you need to quickly process it and decide which to move to storage.

We use 3 different kinds of memory in this process.

  • Sensory memory
  • Working memory
  • Long-term memory

Sensory memory takes in the visual data we receive during fixations. The visual impression that reaches memory won’t last long so we buffer it. This allows some to remain long enough to pass into the next stage, working memory.

The greater the cognitive load, the more difficult it is to process information

Working memory helps us make sense of the visual world in front of us. New data is analyzed and synthesized with what’s come before. We do conscious mental work to decide how to integrate new information with old information and we use both to better understand what we see and determine where to focus next.

Like sensory memory, working memory has limited capacity and data can only last in it for a short time. Data in working memory decays rapidly unless it’s repeated and held long enough to reach long-term memory. Chunks of information (about 3–5 at a time) are best for processing.

Working memory is where we map our surroundings. It’s where we separate figure from ground. It’s where we identify shapes by searching long-term memory for anything similar to the shapes our senses take in.

It might surprise you to know that long-term memory is capable of storing unlimited information. While we can store it all permanently, what we can’t do as well is retrieve the information when we want. Since, we’re not consciously aware of what’s there or where it might be located it’s hard to find.

To help we organize long-term memory into 3 different types of memory where we store different things in different ways.

  • Semantic memory — is interpretation. It stores meaning, facts and concepts
  • Episodic memory — is autobiographical. It stores events and associated emotions relating to the experience
  • Procedural memory — is task oriented. It stores skills and procedures and other information about how to do things.

The more meaning we give to any information, the more we include both visual and verbal information and connect images to words, the greater the depth of processing for that data. The greater that depth, the more likely the information will be stored deeper in long-term memory.

Cognitive Load

To search memory for associations with what we see takes effort. To process information and perform simple tasks like counting or more complex tasks like problem solving requires effort. Just deciding where to fixate next requires effort.

This effort is called cognitive load. The more we ask of working memory, the greater the cognitive load. The greater the load, the more difficult it is to process information and the more likely any information will be misinterpreted, misunderstood, or overlooked entirely.

One goal for designers is to reduce cognitive load.


To make things easier to find and understand, long-term memory organizes itself further into schemas. Schemas are mental constructions that help us classify and categorize the information we store. They’re abstract and general representations of our visual world.

For example the schema for a car might include information about its general shape and knowledge that a car has 4 wheels, an engine and a steering wheel. It might include information that cars transport people from one location to another.

When we see something with 4 tires and an engine that’s in the general shape of a car our schema helps us determine if the thing we see is or isn’t a car.

Schemas are dynamic. As new information is taken in they refine themselves with the new information. The new is assimilated into the stored information and the process is one of learning.

Our mental models go further and help explain cause and effect and how changes to one object affect another. It’s also how two people can view the same events and walk away with such a completely different interpretation of those events. Each had a different mental model for how cause and effect works.

Moving the Eye with Your Design

With the bottom-up process we looked at how designers might create contrast to attract the eye’s attention and influence where it moves next. We can work the other direction as well and influence the eye through top-down processing.

For one we can consider the cognitive characteristics of our audience and design based on where our visitors might lie along the scale of each characteristic.

  • Developmental — What experience, skill, and understanding of the subject will the audience have?
  • Distractibility — What is their ability to focus?
  • Visual literacy — How much knowledge and understanding of graphic symbols and techniques do they have?
  • Motivation — How interested will the visitor be? Does the information present match his or her goals?
  • Culture — How does the culture of the viewer provide context for visual input in your design?

Each of the above requires knowing your audience as best you can and aligning content and visuals with their interest, culture, and experience.

Another thing designers can do is to invoke the most appropriate mental process for the information being conveyed. For example in the case of an event, a new product, or an announcement the main purpose might be to get noticed.

In that case your goal would be to attract the viewer’s eye and hold the eye’s attention long enough for the information to be taken in and processed. You could contrast the feature channels we talked about last week to attract that attention.

If the information exists to enhance knowledge and understanding you might decide to make greater use of maps and infographics. These can make things clearer and reveal relationships with previously stored information. Ideally you’d create these information graphics to be well organized and easy to interpret.

Getting visitors to use the most appropriate mental process requires deep understanding of the information you’re delivering and it’s purpose as well as the different kinds of mental processing human beings do.


Information doesn’t just find it’s way into long-term memory. A lot of work and processing has to occur before information is stored for later use.

Data passes through different kinds of memory and is stored differently based on the type of information it is. It’s further organized in long-term memory into schemas to help us classify and categorize what we see.

What’s stored in long-term memory is used to identify and understand new visual data. Long-term storage shapes our interpretation of new data and helps us decide where to look next.

While it takes more effort than simply contrasting visual features, we can take advantage of the top-down process by understanding our visitors and our information and seeking to reduce cognitive load always and everywhere.

Download a free sample from my book, Design Fundamentals.

Leave a Reply

Your email address will not be published.