CRI Meaning Lighting: What the Score Measures and Where It Fails
CRI is a score out of 100 for how faithfully a lamp shows the colour of an object. The comparison is against a reference light of the same tint. It is not brightness, and it is not the colour of the light itself. The figure most products print is an average of eight pastel test colours. That is why a lamp can score 90 and still make red fabric look dull.
That single sentence explains most of the arguments I see over lamp samples.
A buyer compares two lamps at 90, chooses the cheaper one, and finds that meat, wood and skin all look grey in the finished installation. Nothing was mislabelled. The index simply does not measure the colour that failed.
This page covers what the number is built from and where the red gap comes from. It also covers why the CIE’s own committee called the method flawed. Then what to write on an order instead.
The Short Answer
A lamp with a high CRI shows colours close to how a natural light of the same tint would show them. A low score means some colours shift.
The scale runs to 100, and only a source whose spectrum matches daylight or a black body reaches it. It can also go below zero: low-pressure sodium lamps score negative.
Two figures are usually quoted. CRI is the general name, and CIE Ra is the actual standard value printed on boxes. When a datasheet says “CRI 90”, it means Ra of about 90.
What the number does not tell you is the tint. A 2,700 K lamp and a 6,500 K lamp can both score 90, and they will look nothing alike. Tint is a separate measurement.
What the Number Compares Against
Colour rendering is a comparison, not an absolute. The test asks how a set of coloured samples looks under the lamp. The same samples are then shown under a reference light.
The reference is chosen to match the lamp’s own tint. Under 5,000 K the reference is a black body at the same temperature. At or above 5,000 K it is a phase of standard daylight.
That choice is what makes the score fair across tints. A warm lamp is compared with a warm reference, so it is not punished for being warm.
The two lights are also adjusted for how the eye adapts to a tint. That keeps the comparison like for like. Without that step, every warm lamp would fail against a cool reference.
There is one more condition, and it is the one buyers never hear about. The method only applies to sources that are close to white. The colour difference between the lamp and its reference must stay under 5.4 thousandths in the standard colour space.
Outside that limit the score is not defined. A strongly tinted lamp can be measured, but the answer does not mean what the box implies.
How Ra Is Built: Eight Swatches and a Mean
The standard behind the number is CIE 13.3, published in its common form in 1974. It uses fourteen coloured samples taken from an early Munsell atlas.
Each sample is lit by the lamp and by the reference in turn, and the colour difference is measured. A sample that shifts a lot loses points, on a straight line. The score is 100 minus 4.6 times the measured difference.
Then the average is taken. That average is Ra, and it is where the trouble starts. Only the first eight samples go into it.
| Sample | Colour under daylight | Counted in Ra |
|---|---|---|
| R1 | Light greyish red | Yes |
| R2 | Dark greyish yellow | Yes |
| R3 | Strong yellow green | Yes |
| R4 | Moderate yellowish green | Yes |
| R5 | Light bluish green | Yes |
| R6 | Light blue | Yes |
| R7 | Light violet | Yes |
| R8 | Light reddish purple | Yes |
| R9 | Strong red | No |
| R10 | Strong yellow | No |
| R11 | Strong green | No |
| R12 | Strong blue | No |
| R13 | Light yellowish pink, a skin tone | No |
| R14 | Moderate olive green, a leaf tone | No |
Read the first eight again. They are deliberately low in saturation and spread evenly around the hue circle. That was a reasonable choice in 1974, when the aim was a fair average.
It also means the average is built from pastel colours. The six saturated samples and the two skin and leaf references are outside the sum. So is everything a customer actually notices.
The Red That Falls Outside the Average
R9 is the strong red sample, and it is the one that matters most in practice.
It sits outside the average by design. A lamp can therefore post a respectable Ra while scoring badly on R9. Any supplier that measures properly reports the two figures separately.
This is not a rare edge case. Warm white light comes from putting a phosphor over a blue emitter. The red end of the spectrum is exactly where that approach gives least. A dip in the red region is the usual reason a high-Ra lamp still looks flat on meat, timber or skin.
The visible result is consistent. Reds pull toward brown, skin loses warmth, and wooden surfaces lose their distinction from one another.
So the practical question for a buyer is not “what is the Ra”. It is “what is the Ra, and what is the R9”.
An order that names only the first figure leaves the second one to the factory’s cheapest phosphor blend.
You can see the effect without instruments. Put a red object and a white sheet of paper under the lamp at the same time. If the paper stays neutral and the red goes brown, the red score is the one that is short.
Do the same with a wooden surface and a piece of grey card. Timber has a wide spread of warm tones. A lamp with a thin red region flattens the grain into one shade, and the material loses its character.
An Index Its Own Authors Call Flawed
This is the part that surprises buyers who assume a standard number is beyond argument.
The reference literature describes the 1974 CRI as acknowledged to be flawed. It is still widely used for consumer lighting. An update was published in 1999, and it is rarely followed.
The criticism is specific. The index is described as a poor guide for visual assessment, especially below 5,000 kelvin. That range covers most domestic and retail lighting.
That is not a maker’s complaint. It is the reason a newer method was developed and formally endorsed.
None of this makes the number useless. It makes it a floor rather than a promise. A high Ra tells you the pastel range is handled. It says nothing about the reds, and nothing about whether the light makes a room look good.
TM-30: Two Numbers Instead of One
A replacement exists, and it has been endorsed rather than merely proposed.
The method is IES TM-30, published in 2015 and revised in 2020. It is described as the current recommended measure for colour rendering as people actually perceive it. The CIE endorsed it in 2015 to replace CRI.
It uses 99 colour samples instead of fourteen, spread across the whole colour space rather than a pastel ring. Measurements are made in a newer colour space built around how vision behaves.
It also reports two headline numbers instead of one. Rf is fidelity, on the same 0 to 100 idea as Ra: how close the colours land to the reference. Rg is gamut, which describes whether colours are dulled or stretched.
It also breaks the result down by hue. The output includes a score for each of sixteen hue ranges, plus a chart of the overall gamut shape. A designer can then see which colours are pushed and which are pulled in.
That second number is the real departure. A lamp can score under 100 on fidelity and still make skin look better, because a little extra red is often preferred to exact accuracy.
TM-30 formalises that trade. The standard sets out design intents that allow fidelity to be given up on purpose. A single averaged score can never express that.
The fidelity figure has since been adopted as a CIE standard under the name colour fidelity index. So a supplier reporting Rf is not using a fringe metric.
| Measure | Samples | What it answers |
|---|---|---|
| CIE Ra, 1974 | 8 of 14, pastels only | Is the pastel range close to the reference |
| R9 on its own | 1 strong red | Does red fabric, meat or timber hold up |
| TM-30 fidelity, Rf | 99, whole space | How close colours land overall |
| TM-30 gamut, Rg | 99, whole space | Are colours dulled or stretched |
What the Scores Look Like
For context, the published ranges run from below zero to a full 100.
Low-pressure sodium, the orange street light of old photographs, scores negative. Basic fluorescent tubes sit near 50. The best multi-phosphor fluorescent types reach about 98.
Typical white LEDs are quoted at 80 or above, and some makers claim up to 98. The maximum of 100 belongs to a source whose spectrum matches the reference almost exactly. In practice that means an incandescent lamp or a very good daylight simulator.
That spread is worth holding in mind. A claim of 98 on a low-cost LED is not impossible. It is the upper end of the market rather than the middle, and it should come with a test report.
Six Lines for the Order
A single Ra figure is the weakest useful form of this specification. These six lines make it hold.
- Ra, with the method. Write the Ra value and name the standard it was measured to, at a stated colour temperature. The score is tied to that temperature.
- R9 on its own, with a minimum. This is the line that protects reds. Without it the average can be met while the red sample collapses.
- R13 and R14 where they matter. Skin and leaf samples, for anything sold into food, retail or hospitality.
- Rf and Rg if the supplier can report them. Ask for both. A fidelity figure without a gamut figure hides the trade the designer made.
- Measurement condition. Name the observer, the drive current and the case temperature. A rendering figure measured cold does not describe a warm fixture.
- Sample against production. State the lot or bin and how far the score may fall between the approval sample and the shipment.
Line two is the one I would never leave out. Everything else can be argued after the fact; the red score cannot be recovered once the phosphor is chosen.
Where the product is a work light, the rendering figure and the tint figure are usually written together. The work light order sheet carries both on one line.
For ceiling and task lighting in a workshop, the rendering question often comes second to the tint question. If that is where the doubt is, start with the colour temperature guide, then come back for the fidelity figure.
How to Compare Two Quotes
Three checks settle most of these comparisons without a laboratory.
First, look for R9 next to the Ra. If the quote names one figure only, the supplier has reported the part of the standard that flatters the product.
Second, check the colour temperature the score was measured at. A lamp quoted at 90 at 4,000 K is not the same product at 2,700 K, and the rendering figure will move with it.
Third, ask whether the figure came from the finished product or from the bare emitter. A chip measured on a bench and a lamp measured in its housing are different claims.
After those three, price and warranty belong in the conversation. A cheaper lamp that meets Ra and misses R9 is not cheaper in a shop where meat is the product.
Brightness claims have their own traps, and they are separate from this one. The lumens and watts comparison covers why a watt figure stopped meaning output. Neither number tells you the other.
Where a claim does end up being tested is in the paperwork. The safety and emissions file is the same file that carries the photometric report, and it is the one a customer will ask for.
One Missing Figure Sinks a Red Display
Send me the application, the colour temperature and the rendering figures on the quote. I will tell you which of those figures the order needs. I will also point out where the red score was left out, and what to write so the sample and the shipment match. That costs nothing before tooling.
Send the Rendering SpecWhat counts as a good CRI?
Household products usually claim 80 or more, and 90 is common in better lamps. But the average covers pastel colours only, so a high figure needs an R9 value beside it before it means much.
Is CRI the same as colour temperature?
No. CRI describes how faithfully colours are shown. Colour temperature describes the tint of the light itself. A warm lamp and a cool lamp can both score 90.
Can CRI go above 100?
No. The maximum is 100, reached only by a source whose spectrum matches the reference. The scale can go below zero, and low-pressure sodium lamps score negative.
What is R9 and why does it matter?
R9 is the strong red sample. It is excluded from the Ra average, so a lamp can post a good Ra while showing reds poorly. Red is where warm white LEDs and basic fluorescents usually fall short.
What is the difference between Ra and Rf?
Both are fidelity scores out of 100. Ra averages eight pastel samples under a 1974 method. Rf covers 99 samples across the whole colour space and comes with a second figure for gamut.
Do I need TM-30 figures on an order?
Only if colour is the product, such as retail displays, food, textiles or studio work. For general lighting, a measured Ra with a stated R9 is usually enough to hold a supplier to.
Questions & Comments
We read every one and reply within 24 hoursThe Ra figure is an average of eight pastel samples. Red is not one of them. Tell me what the product lights, and I will say which rendering figures the order needs.