Design of everyday things

The Design of Everyday Things is Don Norman’s argument that usable products make the right actions discoverable, understandable, and recoverable. When many people repeatedly make the same error, the first question should not be what is wrong with the users. It should be what the design asked them to infer, remember, notice, or control.

Good design shifts work from the user’s memory into the environment. Signifiers show where to act, mappings make controls correspond to outcomes, constraints prevent invalid actions, feedback reveals what happened, and a coherent conceptual model lets the user predict what will happen next. A good design makes the correct action easier to infer than the incorrect one.

Core framework

• Discoverability: The possible actions and current state of a system should be easy to determine.

• Affordance: The relationship between an object and an actor that makes a class of actions possible. A handle affords pulling because of its physical form and the capabilities of the person using it.

• Signifier: A perceivable cue that communicates where or how an action should occur. Affordances determine what is possible — signifiers communicate what is intended.

• Mapping: The relationship between controls and their effects. Natural mappings reduce the amount a user must memorize.

• Feedback: Perceivable information about the result of an action and the new state of the system.

• Conceptual model: The user’s simplified explanation of how the system works and how actions produce outcomes.

• Constraint: A limitation that narrows the set of possible actions. Constraints can be physical, logical, semantic, or cultural.

• Gulf of execution: The gap between what a user wants to do and their ability to determine how to do it.

• Gulf of evaluation: The gap between what a system did and the user’s ability to understand the result.

• Slip: The goal is correct but the action is executed incorrectly.

• Mistake: The action follows the user’s plan, but the plan itself is wrong because the situation or system was misunderstood.

Discoverability and understanding

A device can be technically functional and still fail if the user cannot determine what actions are available or what state the system is in. Norman separates two requirements: discoverability — figuring out what can be done — and understanding — forming a usable model of how the system works.

A door that must be pushed should not need a sign saying PUSH. Its shape, hardware, and surrounding cues should make pushing the obvious action. A stove with four burners and four controls should make the relationship between each knob and burner immediately visible. A digital control should reveal whether an action succeeded, failed, or is still processing.

The recurring design failure is to place essential information inside the designer’s head instead of in the product. Designers know how their systems work and therefore underestimate how much invisible knowledge the interface requires from a first-time user.

Key idea: A system is not understandable because its designer can explain it — it is understandable when the system itself supplies enough information for the user to form the right model.

Affordances and signifiers

An affordance is an action possibility. A chair affords sitting, a button affords pressing, and a horizontal bar may afford pushing. But an affordance can exist without being obvious. A hidden gesture in software may be possible while remaining undiscoverable.

That is why Norman distinguishes affordances from signifiers. Signifiers are the visible, audible, tactile, or conventional cues that tell the user where action is expected: a raised button, an underline indicating a link, a grip texture, a label, a shadow, a cursor change, or an empty field with a prompt.

Affordances determine what can be done; signifiers determine whether the user can tell.

This distinction matters because designers often solve the wrong problem. Adding capability does not improve usability if the capability remains invisible. Conversely, a strong signifier can communicate an interaction even when the physical affordance is weak, as in touchscreens where every region of glass has essentially the same physical properties.

Key idea: The action must not merely be possible — the intended action must be perceivable.

Mapping and conceptual models

Mapping describes how controls correspond to outcomes. The best mappings exploit spatial, physical, or cultural relationships that already make sense. A steering wheel turns in the direction the vehicle turns. A vertical volume control that moves upward to increase volume uses a familiar spatial metaphor. Stove controls arranged in the same pattern as the burners remove the need to decode arbitrary labels.

Poor mappings impose memory. When four identical switches are placed in a row while controlling objects arranged in a square, users must memorize relationships the product could have represented directly.

Mappings feed a broader conceptual model. A person does not need to understand the internal engineering of a system, but they need a model accurate enough to predict its behavior. The refrigerator can be understood as something that becomes colder when the temperature setting is lowered without knowing the thermodynamics of refrigeration.

The user constructs this model from the system image — everything the product exposes through its form, controls, labels, documentation, conventions, and feedback. If the system image does not reflect the designer’s intended model, the user will construct a different one.

Key idea: Good design communicates causality through the product itself rather than requiring the user to reconstruct it from instructions.

Feedback and the two gulfs

Every interaction begins with a goal and ends with an evaluation. Norman models action as a cycle: the user forms a goal, decides what to do, executes an action, perceives the result, interprets it, and decides whether the goal was achieved.

Failures can occur on either side. The gulf of execution appears when a user knows what they want but cannot determine what action the system requires. The gulf of evaluation appears when the user acts but cannot tell what happened.

Feedback closes the second gulf. A physical switch changes position. A button changes state. A file upload displays progress. A payment confirms completion. Good feedback is immediate enough to preserve the perceived relationship between cause and effect, but informative enough to communicate the new state rather than merely acknowledge the click.

A system should make both the next action and the consequence of the previous action legible.

Key idea: Interaction fails when either action or interpretation requires more inference than the interface supplies.

Human error is often a design problem

Norman distinguishes slips from mistakes. A slip occurs when the intended goal is correct but execution fails — pressing the adjacent button, typing the wrong character, or performing a familiar action automatically in the wrong context. A mistake occurs when the user forms the wrong goal or plan because their understanding of the system is wrong.

Both categories imply different remedies. Slips call for larger targets, clearer modes, confirmations for destructive actions, undo, sensible defaults, and constraints that make accidental actions difficult. Mistakes call for better conceptual models, clearer information, better signifiers, and less ambiguous system behavior.

“When people err, change the system so that type of error will be reduced or eliminated.”
— Don Norman, The Design of Everyday Things

The deeper principle is that repeated human error is data. If many competent people fail in the same way, blaming attention or intelligence explains little. The design has created a predictable failure mode.

Key idea: Design for the person who will occasionally be distracted, rushed, inexperienced, or wrong — not for an ideal user who never deviates.

Constraints and forcing functions

Constraints reduce the space of possible actions. A key fits only one orientation. A connector shape prevents incompatible insertion. A grayed-out command communicates that an operation is unavailable. These restrictions can improve usability because they convert uncertainty into structure.

Norman identifies several kinds:

• Physical constraints make an action mechanically impossible.

• Logical constraints use relationships within the system to rule out alternatives.

• Semantic constraints rely on knowledge of the situation and what actions make sense.

• Cultural constraints rely on learned conventions shared by a group.

A stronger form is the forcing function, which requires one condition before another action can occur. An interlock prevents a step until a prerequisite is satisfied. A lock-in prevents premature exit from a process. A lockout prevents entry into a dangerous state.

Constraints are most valuable when the cost of error is high and the valid action space is narrow. They replace instructions with structure.

Key idea: The safest interface often prevents the error rather than warning about it afterward.

Knowledge in the head and in the world

People can operate complex systems using two sources of knowledge: knowledge in the head and knowledge in the world.

Knowledge in the head includes memory, skills, learned conventions, and internal models. It can be fast and efficient once mastered, but it imposes learning costs and is vulnerable to forgetting. Knowledge in the world includes labels, visible controls, examples, signs, menus, spatial arrangements, and other external cues.

Good design chooses deliberately between them. Frequently repeated expert actions can justify learned shortcuts. Infrequent, high-stakes, or unfamiliar actions should usually expose more information in the environment. A fire exit should not require recall. A destructive command should not depend on remembering an undocumented shortcut.

Standards and conventions are especially powerful because they move knowledge from each individual product into shared culture. Once users learn that underlined text is clickable or a magnifying glass means search, every compliant product becomes easier to learn.

Key idea: Memory should be reserved for knowledge worth learning; arbitrary interface details should be externalized whenever possible.

Human-centered design

Norman’s design process is iterative rather than deductive. Designers rarely begin with a complete understanding of the real problem. They observe people, identify unmet needs, generate possibilities, prototype, test, and revise.

The essential move is to distinguish the stated request from the underlying need. Users are often good at revealing where friction occurs but less reliable at specifying the optimal solution. A request for a larger button may actually indicate that the hierarchy is unclear. A request for more settings may indicate that defaults are poor.

Human-centered design therefore uses behavior as evidence. Watch what people attempt, where they hesitate, which errors repeat, and what workarounds they invent. Prototypes turn assumptions into observable interactions cheaply enough to revise before implementation becomes expensive.

Key idea: The designer’s job is not to obey every requested feature — it is to discover the human problem the request is evidence for.

Implications

The book’s durable contribution is to relocate responsibility. Confusion is not automatically a property of the user. It can be engineered into a product by weak mappings, invisible state, ambiguous signifiers, poor feedback, arbitrary conventions, and missing constraints.

This does not mean every system must be simple. Some domains are irreducibly complex. The design problem is to make necessary complexity legible while eliminating unnecessary complexity created by the interface itself.

The practical test is straightforward:

• Can the user tell what actions are possible?

• Can they tell which action is appropriate?

• Does the relationship between control and result make sense?

• Can they tell what happened after acting?

• Can common errors be prevented or reversed?

• Does the product teach a coherent model of itself?

When a system repeatedly requires explanation, memory, or recovery from predictable mistakes, part of the intelligence belongs in the design rather than in the user.

Key idea: Good design makes intention, action, causality, and system state visible enough that ordinary human behavior produces successful use.