The Norwood scale: what the stages actually describe

The Norwood scale is a classification system. Its job is to take the enormous variety of ways a scalp can look and compress it into a handful of named categories, so that two people describing the same head land on the same word. That is the whole design goal. It was built for communication, first between clinicians and later between researchers who needed a way to group participants consistently across different studies and different countries.

It has a lineage worth knowing, because the lineage explains the shape. James Hamilton published a classification of male pattern hair loss in the early 1950s. O'Tar Norwood revised and extended it in the mid-1970s, adding categories that Hamilton's original did not separate cleanly. That is why the scale is properly called the Hamilton-Norwood scale, and why the numbering has the slightly uneven feel of something that was amended rather than designed from scratch. It was never intended to be a self-service quiz on a phone.

The most important structural fact about it is what it leaves out. There is no time in a Norwood classification. There is no hair count. There is no measurement of shaft thickness. There is no statement about cause. There is a drawing of a head, and a question about which drawing yours most resembles right now. Everything people wish the scale told them lives outside the scale.

What follows is a plain description of what each category depicts. Read them as portraits, not as steps on a staircase you are obliged to climb.

Stage I

A hairline with no meaningful recession. This is the reference state the rest of the scale is described against. Worth noting: a completely straight, low, juvenile hairline is a childhood feature, and a modest maturing of the hairline at the temples in early adulthood is an ordinary developmental change rather than the start of a pattern. The scale does not draw that distinction for you, which is one of the reasons self-assessment goes sideways early.

Stage II

Symmetrical triangular recession at the temples, leaving the central forelock forward of the temporal points. The frontal hairline has taken on a shallow V or M shape rather than a straight line. This is a common source of self-assessment error, because it overlaps with the ordinary adult hairline shape described above.

Stage III

Deeper temporal recession, extending further back, with the recessed areas becoming sparse or bare rather than merely thinner. Norwood treated this as the threshold at which the pattern is unambiguous rather than arguable. Stage III also carries a specific variant, described below, in which the loss at the crown rather than the temples is the defining feature.

Stage IV

Frontal and temporal recession that is more severe than stage III, combined with a distinct area of loss at the vertex — the crown. Critically, a band of hair still separates the two regions across the top of the head. That bridge is the defining feature of the stage, and its presence or absence is what distinguishes stage IV from what follows.

Stage V

The same two regions, but the band between them has narrowed and thinned. The two areas are still separate, and the connecting bridge is still there, but it is no longer robust. In practice this stage is defined more by the state of the bridge than by the size of either bald region.

Stage VI

The bridge is gone. The frontal and vertex areas have joined into a single continuous region, and hair remains around the sides and back. The horseshoe shape most people picture when they think about pattern hair loss is this stage and the next one.

Stage VII

The most extensive category. A narrow band of hair remains on the sides and the back of the head, running low around the skull, with the entire top area bare. There is no stage VIII; VII is the terminal label in the system, which means the scale stops describing differences past that point.

Two lettered qualifiers get attached to the numbers, and both exist because the plain numbered series could not describe certain real heads.

The vertex variant covers a presentation where loss at the crown is the leading feature and the frontal hairline shows relatively little recession. It is attached to stage III because Norwood needed a way to say "this person has a defined bald area at the crown but a frontal hairline that has not moved much," and the numbered series alone forced that person into a category that misdescribed them.

The A variants describe a different shape entirely. Instead of temples receding faster than the middle and leaving a forelock, the entire frontal hairline retreats backwards as a fairly even front, without a distinct island of hair left in the centre and without a separate crown patch developing on its own. The result is a broad advancing border rather than the classic M. A person following the A pattern can look quite different from a person at the equivalent plain-numbered stage even though both are correctly labelled.

Both variants make the same point about the scale as a whole. It is a set of buckets, and buckets get added when the existing buckets fail to describe something. That is a sign of a workable descriptive system, and it is also a reminder that the categories are conventions rather than natural kinds. Nothing in biology says there are exactly seven states.

Here is the core misunderstanding. The stages are ordered by extent, so they look like a sequence in time. They are not one. The scale contains no rate, no duration, no arrow. A stage is a snapshot label: this is what this head looks like today. Reading it as a position on a conveyor belt adds a claim the system never made.

The practical consequence is that a stage cannot answer the question people most want answered. "I am a stage III, so how long until I am a stage V" is not a question the classification has any means of addressing, because the classification has no memory of how you arrived and no model of what happens next. Two people who look identical today may have arrived over two years or over fifteen, and that difference — which the label discards entirely — is exactly the information a prediction would need.

A second consequence: the scale is not a diagnosis. It says nothing about cause. A visible pattern is a pattern, and there are causes of hair loss that are not androgenetic, some of which look superficially similar in a photograph and are managed completely differently. A label describing a shape cannot distinguish between mechanisms that produce similar shapes. That is a clinical question and it requires an examination, not a chart.

There is a third thing the scale is not, and it is the one that generates the most wasted worry: it is not a severity ranking of how you look to other people. It ranks extent of a pattern. Whether extent maps to how anyone perceives you involves face shape, hair colour and contrast against scalp, how the remaining hair is cut, and a great many things that a category label has no access to.

None of the above makes the scale worthless. It makes it a different tool than people think. Shared vocabulary has real value. A clinician writing notes, a surgeon planning coverage, a researcher grouping participants — all of them need a compact way to say what a head looks like without attaching a paragraph of prose. The scale does that job, and it does it well enough to have survived for decades.

It is also useful as a change-tracking frame, provided the comparison is done honestly. If the same person is photographed under the same conditions a year apart, the category is one of several ways to describe what moved. The value there is in the controlled comparison, not in the label. The label is the summary; the discipline of the capture is what makes the summary mean anything.

What it is bad at is self-assessment from casual photographs, which is how it usually gets used online. The rest of this article is about why.

The scale asks you to compare your head to a picture. That sounds simple until you notice that you are comparing one picture, taken under uncontrolled conditions, against a stylised drawing made under no conditions at all. Apparent hair density in a photograph is not a property of the hair. It is a property of the hair, the light, the camera position, the hair length, whether the hair is wet, how it was styled, and what is behind it.

This matters more for hair than for almost any other appearance feature, because what you are actually judging is the contrast between hair and the scalp visible between it. Hair is thin and separated; scalp shows through gaps. Anything that changes how much light falls into those gaps, or how much the strands overlap, changes the perceived density directly. The mechanism is optical, and it is large.

The sections that follow break the variables apart. Every one of them can move an apparent stage on its own.

Overhead light is the classic offender. A ceiling fixture, or the sun near midday, throws light down onto the top of the head. Where hair is dense, that light is caught and scattered by the strands. Where hair is thinner, light passes through and reaches scalp, which is a broad, smooth, reflective surface. The result is that a thinning crown lights up under overhead illumination in a way it simply does not under front-on light. Bathroom mirrors are usually lit from above. So are most public toilets, most lifts, and most of the places people accidentally catch themselves and panic.

Direct flash close to the lens does something related but distinct. It produces hard specular highlights on the scalp and flattens the shadow that would otherwise sit between strands, so the scalp reads brighter and the hair reads sparser. Soft, diffuse, frontal light does the opposite: it fills the gaps evenly and hair looks thicker. Neither is the truth. They are two renderings of the same head, which is why a comparison made across different lighting tells you about the lighting rather than about the hair.

Background matters too. A dark-haired person photographed against a bright window has their hairline eaten by flare and their hair edge rendered as a thin fringe; the same person against a mid-toned wall reads denser. Human perception of density is relational, exactly as it is for tooth colour — a point we make in detail in the before-and-after photo guide, where the same optics problem produces the same false certainty.

Wet hair clumps. Surface tension pulls adjacent strands into bundles, and bundles leave gaps between them that dry hair does not have. A head that reads as full when dry can read as visibly sparse when wet, with the scalp clearly showing along the part and across the crown. Nothing has been lost. The strands have simply rearranged themselves from an even distribution into a smaller number of thicker clusters.

Wet hair is also darker and lies flatter against the skull, which removes the volume that lifted strands away from the scalp and reduces the apparent distance between hairline and skin. Combine wet clumping with the overhead bathroom light described above, and you have the single most alarming and least informative view of your own head available anywhere. It is also the view most people use, because it is the one they see every day.

The rule that follows is simple and worth applying strictly: never compare a wet head to a dry one, and never assess a stage from a shower mirror. If you want to track wet hair, track wet hair against wet hair. Consistency is what makes a comparison legitimate, not which state you pick.

Longer hair drapes. It falls across neighbouring areas, overlaps, and covers scalp that the same hair would not cover if it were shorter. That is not concealment in any dishonest sense; it is simply what longer strands do under gravity. It means, though, that hair length is a confound in every before-and-after comparison of density, and one that people almost never control for.

Cutting hair short does the reverse. Shorter strands stand more upright, cover less area each, and reveal the underlying distribution more faithfully. People who have just had a short cut frequently believe they have lost a great deal of hair in a week. What actually changed is how much scalp is exposed to view per strand. Buzzing shorter still can flip the effect again, because at very short lengths the contrast between hair and scalp drops and the pattern can become less visible rather than more.

Product changes the same variable deliberately. Anything that stiffens or separates strands alters how they sit relative to each other, and therefore how much scalp shows through. Where the part is placed moves the single most scalp-revealing line on the whole head. None of this is trickery; it is the everyday physics of how a dense field of thin strands arranges itself over a curved surface. But it does mean that a stage assessed on a styled head and a stage assessed on an unstyled head are not the same measurement.

A hairline photographed from below reads lower and denser, because you are seeing the front edge nearly edge-on and the recessed areas behind it are hidden by the forelock. The same hairline photographed from above reads higher and thinner, because you are now looking down onto the top of the head where the recession lives, and the crown enters the frame as well. A phone held at chest height and a phone held above eye level produce two different apparent stages of the same head.

Head tilt does the same thing with less warning, because most people do not notice they are doing it. Tipping the chin down brings the crown into view; lifting it hides the crown and exposes the frontal hairline. In a casual photograph taken by someone else, you have no control over either of these, which is why candid shots from parties and group photos are the source of so many sudden discoveries.

Lens and distance add a further layer. A phone camera held close applies visible perspective distortion, enlarging whatever is nearest the lens and compressing what is further away. On a forehead-forward angle that exaggerates the forehead and hairline relative to the rest of the head. The same distortion is what makes facial proportions unreliable in close selfies — the reason our own 17-metric scan and jawline test both ask for a level camera at a sensible distance before they ask for anything else.

We can offer a number here from our own measurements, and it is worth stating precisely because it is about the general problem rather than about hair specifically. We compared scans from the same person taken on different calendar days less than fourteen days apart — 103 people who scanned the same face twice. The median difference between the two Face Score percentiles was 4 points, and 40 percent of them moved 5 points or more.

That is a facial-geometry result, not a hair result, and we are not extending it to hair density directly. The reason it belongs in this article is the inference it licenses. Facial geometry is bone and soft tissue. It is about as stable as an appearance feature gets, and it does not meaningfully change in under two weeks. The score moved anyway, because the photographs differed — light, angle, distance, expression — and the measurement reads the photograph rather than the person.

If a stable, rigid feature is that sensitive to capture conditions, a feature made of thousands of thin, movable, wettable, cuttable strands judged by eye against a drawing is not going to be more reliable. It is going to be considerably less. That is the honest bound on any Norwood stage you assign yourself from a phone photo.

This is the second big misreading, and it costs people money. The idea that a stage maps onto a treatment — that stage II means one thing, stage V means another — treats a descriptive label as a decision rule. It is not one, and no serious clinician uses it that way.

What actually enters the decision is a much longer list. Whether the loss is androgenetic at all, since several other causes exist and some are reversible or require treating something else entirely. How old you are and how long the pattern has been developing. Whether it is currently changing, and how fast, which requires two controlled observations rather than one. The density of the areas that still have hair, including the areas a surgeon would take from, which is a different question from how much is missing. Scalp condition. Other medical history and other medications. What you actually want, which is not always more hair and is sometimes just a clear answer.

Two people who present at the same visible stage can reasonably be told very different things, because the inputs above differ. The stage is where the conversation starts, not where it ends. Anyone who quotes you a plan from a stage number and a phone photo has skipped every input that matters.

Two drug names come up constantly in this conversation, and we will describe what they are and nothing more. Minoxidil is a topical or oral agent originally developed for another purpose; its effect on hair is understood to involve the vascular and growth-cycle environment of the follicle rather than hormones. Finasteride is an oral drug that acts on the enzyme pathway converting testosterone to dihydrotestosterone, the androgen implicated in patterned follicular miniaturisation.

That is the extent of what belongs in an article. Both are drugs. In many countries at least one of them is prescription-only, both have recognised side-effect profiles, and both involve tradeoffs that depend on your individual medical situation. We are not going to give you a strength, a frequency, a duration, or an expectation, because doing so would be practising medicine badly at a distance from someone we have never examined.

Take it to a doctor, and take these questions with you: is my hair loss patterned or is something else going on, and how do you know? What are the realistic outcomes for someone in my situation, and what would make them less predictable? What are the known side effects, how common are they, and what should make me stop? How will we record a baseline so that any change is measured rather than remembered? That last one is the same discipline this whole article is about.

Surgical hair restoration is often discussed as though a stage number told you whether you are a candidate. It does not, for a reason that is easy to state: the scale describes the area that has lost hair, and the binding constraint in surgery is the area that still has hair. Coverage is a budget problem. What can be moved is finite, and what determines the budget is the density and quality of the region a surgeon would harvest from, not the size of the region they would harvest to.

There is a second reason the stage is insufficient: a static label says nothing about whether the pattern is still changing. Placing hair against a border that is still moving produces a result that ages badly, which is why age, rate of change and family pattern all weigh on the decision alongside the current appearance. Again, these are inputs the classification simply does not contain.

We have written separately about how the two main harvesting techniques differ in the FUE versus FUT comparison, which covers the mechanics and the tradeoffs without recommending either. The relevant point for this article is only that the Norwood stage is one line in a consultation, not the consultation.

You cannot control everything with a phone, but you can hold enough still that the remaining difference carries information. The principle is the one any measurement discipline uses: fix the instrument, fix the conditions, and repeat them exactly.

1. Dry hair, every time. Fully dry, not towel-dry. Pick this state and never compare against a wet one.
2. Same length, or note the length. Photograph immediately before a cut, or immediately after, and be consistent about which. A comparison across a haircut is not a comparison.
3. No product, and the same part. Unstyled, combed the same way, parted in the same place.
4. One light source, from the front, everything else off. A window with the overheads off is ideal. Avoid overhead fixtures and avoid direct flash — those two are the largest single distortions available.
5. Same room, same time of day. Daylight changes across the day and across the year. Mark the spot you stand in.
6. Lock exposure. A long press on the subject locks exposure on most phones. Auto-exposure will happily brighten the scalp for you.
7. Four fixed angles. Straight on at eye level, straight down onto the crown, and both sides in profile. Same order, same distance, same camera every time. Do not switch between front and rear cameras — different sensors, different processing.
8. No filters, no beauty mode, no portrait processing. Export the originals and keep them.
9. Once a quarter, not once a day. Weekly checking measures your mood and the weather, not your hair. Slow features need slow sampling.

Doing this makes one specific thing possible that no amount of staring in a mirror can: it lets you tell whether anything actually changed. That is a different and far more useful question than which drawing you most resemble, and it is the only version of the question that a clinician can do anything with.

Since this is a face-measurement site and the topic invites the assumption, we should be explicit. RealSmile detects 68 facial landmarks in a photograph and produces two outputs from them, in parallel. One is a proportion map: 17 geometric metrics — ratios, angles and distances between landmark points. That map is a description of facial proportion, not a ranking, and we do not blend it into a single quality number. The other, computed separately by a different model, is a validated Face Score percentile.

We keep those two apart deliberately, and we publish why on our research base. When we tested a blended composite built from the geometry metrics against human ratings, it failed — the blended geometry score correlated at roughly r = −0.33, which is to say it pointed the wrong way. The individual measurements are real geometry and they are useful as description. The composite was not a valid predictor, so we killed it. The percentile that ships comes from a separate model validated against averaged human ratings on a public benchmark, and we report what it was tested on rather than asking you to take the number on trust.

What none of that touches is hair. The scan does not measure hair density, hairline position, follicle counts, miniaturisation, scalp condition or skin condition. It cannot assign a Norwood stage and it cannot tell you whether your hairline has moved. It is a facial-geometry instrument, it is limited to facial geometry, and the honest thing to do is say so on the page where someone might reasonably expect otherwise. The same limitation applies to a related question we do answer, eyebrow shape, which we cover in the eyebrow shaping guide: shape is geometry, hair density is not.

• "My stage tells me where I will end up." It does not. There is no time axis in the classification. Extent today says nothing about extent later, and the scale has no mechanism for saying it.
• "I moved a stage in a month." Almost certainly not. You changed hair length, or light, or wetness, or angle. Patterned change on a monthly scale is not what the categories were built to resolve.
• "A higher stage means something is medically worse." The scale is a description of appearance, not of health. It does not measure disease severity and it does not identify cause.
• "My stage decides my treatment." It is one input. Cause, rate, donor density, age, history and preference all enter, and any of them can change the answer entirely.
• "A maturing hairline means I am losing my hair." A modest shift back at the temples in early adulthood is an ordinary developmental change. The scale is not built to distinguish it from early pattern loss, which is precisely why that distinction is a clinical one.

The Norwood scale is a filing system. It is a good one for its purpose, which is letting people describe an appearance to each other efficiently. It becomes harmful the moment it is asked to do three jobs it was never designed for: forecasting, diagnosing, and self-assessment from uncontrolled photographs. Those three misuses account for essentially all of the distress attached to it.

What you can actually do is narrower and more productive. Take a controlled baseline set of photographs, on the terms described above. Repeat it a few times a year. If there is a change, you will have evidence rather than an impression, and you will be able to hand a clinician something more useful than a phone photo from a lift with a ceiling light in it. If there is no change, you will have saved yourself a great deal of unnecessary worry, which is a real outcome even though nobody sells it.

And the general lesson transfers. Any appearance feature judged from a casual photograph is being judged through an instrument nobody calibrated. Control the capture, keep a baseline, and compare against a fixed reference rather than against memory. That is the same argument we make about tooth colour, about facial proportion, and about every number this site produces. It is not exciting advice. It is the only kind that survives contact with the data.