Color Memory Game
By John K.··11 min read

What color is gold: the hex code is #FFD700, the metal is a pale cream

The gold hex code is #FFD700, RGB 255, 215, 0. Polished gold itself, computed from its measured optical constants, is #FFDEA3: a pale cream 6.8 JNDs away and 2.6 times less saturated. Where the deep gold colour really comes from, gold vs yellow, and which hex codes to use for a metallic gold.

The gold hex code is #FFD700. In RGB that is red 255, green 215, blue 0. It is the value every browser draws when you write gold in CSS, and it is the answer to nearly every gold colour code search.

It is also not the colour of gold. I took the measured optical constants of the metal, worked out how much of each wavelength a polished gold surface reflects, and converted that spectrum to a screen colour. The result is #FFDEA3. It is a pale, warm cream, and it looks more like vanilla ice cream than like the swatch on a colour chart.

Gold, the CSS keyword
#FFD700 · L* 86.9, C* 87.2, hue 91.3
Gold, the polished metal
#FFDEA3 · L* 90.2, C* 34.2, hue 81.5

Those two are 15.6 CIEDE2000 units apart, which is 6.8 just noticeable differences. Nearly all of it is saturation. The CSS value carries 2.6 times the chroma of the metal. This piece is about where the extra colour comes from, because it does come from real gold, only not from a single reflection off it.

How the metal was measured

The numbers come from Johnson and Christy’s 1972 table of the refractive index of gold, which is still the standard reference. From the complex index you get the reflectance at each wavelength with the Fresnel equations. I lit that with D65 daylight, ran it through the CIE 1931 observer, and scaled it against a perfect white under the same light. Distances are in CIEDE2000, and one JND is taken as 2.3 units, from our own JND measurement. It is the same pipeline I used for the colour of a mirror.

Two caveats. The figure is for pure, clean, mirror-polished 24 karat gold reflecting a white surround. And the red channel of the true colour is 1.6% brighter than sRGB can show, so #FFDEA3 is clipped by 1.3 units. The L*, chroma and hue above are from the spectrum, not from the clipped hex.

Gold reflects 41% of blue and 95% of red

400 nm40.7%
450 nm40.8%
500 nm47.5%
550 nm79.2%
600 nm91.0%
650 nm95.7%
700 nm97.1%
Share of light that polished gold reflects at each wavelength, head on. Bar colours are only a guide to the part of the spectrum.

Gold is a very good mirror for red, orange and yellow light. Between 600 and 700 nm it reflects 95.1% on average. Below about 500 nm it does much worse: 40.7% between 400 and 480 nm, with a minimum of 40.2% at 471 nm. The step between the two sits in the green. The curve crosses 50% at 504 nm and climbs fastest at 519 nm.

Add it up under daylight and gold reflects 76.8% of the light that lands on it. That is the first reason the measured colour is so pale. The second is that the blue never goes away. A saturated yellow needs the blue channel near zero, and #FFD700 sets it to exactly zero. Gold hands back two fifths of the blue, which leaves the blue channel of the result at 163 out of 255.

Silver, run through the same sums, comes out at #FEFDFD, 0.4 units from neutral. Copper comes out at #F7CFBF, a pale salmon with a hue of 47.7 degrees. Gold is 14.8 units from copper and 19.9 from silver.

Silver, same method
#FEFDFD · reflects 98.5%, 0.4 units from neutral
Copper, same method
#F7CFBF · reflects 68.1%, hue 47.7

Why gold has that step in the green at all is a well known result. In most metals, silver included, the same absorption sits in the ultraviolet where nobody sees it. Gold’s nucleus is heavy enough that its inner electrons move at a large fraction of the speed of light, and the relativistic correction pulls the absorption down into the visible. Pyykkö and Desclaux laid this out in 1979. Without relativity, gold would look like silver.

The deep gold colour is gold reflecting gold

Nobody who has held a ring thinks gold is cream coloured, and they are right. A piece of jewellery is not a flat mirror. It has grooves, engraving, links and an inside surface, and light bounces between gold faces before it reaches you. Every bounce applies the same filter again, so the blue that survived the first reflection at 41% survives two at 17% and three at 7%.

12345
#FFDEA3
C* 34.2, h 81.5
#FFC25E
C* 59.3, h 77.8
#FDAB20
C* 77.3, h 74.9
#F69800
C* 89.9, h 72.7
#EF8700
C* 98.1, h 70.8

The chroma climbs from 34.2 on one reflection to 59.3 on two and 77.3 on three. By the fourth it is 89.9 and already outside sRGB. This is why the inside of a gold cup or a wedding band glows a much deeper orange than the outside, and why a gold bar photographed square on looks washed out next to a pile of coins. The CSS value’s chroma of 87.2 corresponds to something between three and four reflections.

There is a catch, and I think it is the most useful thing in the data. Each reflection also turns the hue toward orange: 81.5 degrees, then 77.8, 74.9, 72.7. CSS gold sits at 91.3 degrees, 9.8 degrees further toward yellow than the metal ever is. So no number of reflections lands on #FFD700. The closest is two, and that is still 12.5 units away, 5.4 JNDs. The keyword called gold is yellower than gold can be.

The web colour that best fits the physics is goldenrod, #DAA520. Its hue is 82.9 degrees, within 1.4 degrees of the metal. It is 7.6 units from a triple reflection, and 5.7 from a double reflection dimmed to about 70% brightness. Wikipedia’s metallic gold, #D4AF37, is 7.8 units from a dimmed double reflection. My opinion: if you want a single flat hex that reads as the metal, goldenrod is a better choice than gold.

Why one hex code cannot be gold

A matte object has a colour you can write down, because it scatters whatever light hits it in all directions. A polished metal reflects its surroundings instead. Its hue is fixed, but its lightness is borrowed from whatever it happens to be facing.

Point the same gold surface at a white wall and you get #FFDEA3. Point it at a mid grey room that returns 20% of the light and you get #7E6B4C, a dull olive brown, with the hue still at 81.5 degrees. That is 5.3 units from #84754E, the screen approximation usually published for Pantone’s metallic gold 871 C, which looks brown on a monitor for exactly this reason. A real gold object shows the whole range at once: a cream highlight, orange where it reflects itself, brown where it reflects the room.

Flat #FFD700
Gradient from the measured stops

So a convincing gold on a screen is a gradient. The stops in the right-hand swatch are all from the measurement: #FFDEA3 for the highlight, #FFC25E for the double reflection, and #BD8E43 and #7D5C29 for that double reflection facing a 50% and a 20% surround. The flat swatch next to it has more chroma than any of them and still looks less like metal.

Angle matters much less than I expected. Tilted to 45 degrees the colour moves from #FFDEA3 to #FFDEA4. It only washes out near grazing incidence, losing a third of its chroma by 75 degrees and two thirds by 85. Warm indoor light shifts it by 4.3 units once your eye has adapted, which is under two JNDs.

Gold vs yellow

On a screen, gold is a yellow pulled toward orange and turned down slightly. CSS gold is 11.5 degrees of hue from pure yellow, #FFFF00, 10.2 L* darker, and about a tenth lower in chroma. The two are 11.6 units apart, 5.0 JNDs, so they are clearly different colours side by side. In shades of yellow I found that every name in the L* 65 to 72 band is a gold or an ochre, which says the same thing from the other side: in everyday naming, gold is where yellow goes when it gets darker.

The metal is further out again. It is 21.3 degrees of hue from pure yellow and 21.0 units away. Physically, gold is closer to a pale orange than to a yellow, and its dominant wavelength is 579 nm, which is on the visible spectrum where yellow starts shading into orange.

What people and paint companies call gold

The xkcd colour survey put gold at #DBB40C. That is 8.9 units from the CSS value, 3.9 JNDs, mostly because it is 12.2 L* darker. It is only 3.4 units from Wikipedia’s metallic gold. So the public and the encyclopedia roughly agree with each other, and both sit darker than the browser.

The open color-names list (version 10.28) has 464 colours with gold or golden in the name. Their median is #D8B160, a soft antique gold at a hue of 84.1 degrees. That is between the metal at 81.5 and the CSS keyword at 91.3, and much nearer the metal.

Median of 464 named golds
#D8B160 · 8.5 units from xkcd gold, 14.3 from CSS gold

Three counts from that list stood out to me.

First, only 2 of the 464 are within one JND of the metal’s single reflection: Bachimitsu Gold, #FDDEA5, and Golden Haze, #FCD896. The colours in the wider list that match it best are called things like Organza, Porcelain Yellow and Corn Stalk. When someone names the actual colour of a gold mirror, the word gold almost never occurs to them.

Second, only 8 are within one JND of #FFD700. The official value is nearly as lonely as the physical one.

Third, 304 of the 464, or 66%, are closer to the metal than to the CSS keyword, and 189 have a hue at or redder than the metal against 98 at or yellower than the keyword. Taken as a crowd, the people who name paints and fabrics have a better eye for gold than the 1980s colour table the web inherited. This is the same kind of drift between a reference hex and its real use that I measured for taupe, mauve and chartreuse, with one difference. Here there is a physical object to check against, and the crowd is the side that is closer to it.

Rose gold, and the colour gold lets through

Rose gold is gold alloyed with copper. I do not have optical constants for the alloys, so I have not computed it, but the two ingredients set limits. Pure gold is at 81.5 degrees of hue and pure copper at 47.7, and a mixture should fall between them. The usual rose gold hex code, #B76E79, has a hue of 11.9 degrees. That is 36 degrees pinker than copper with no gold in it at all. The hex is a cosmetics pink that borrowed a metal’s name. If you want something closer to the alloy, start from the copper value and move toward gold.

One more measurement, because it surprised me. Gold leaf is thin enough to pass a little light, and what gets through is the part gold is worst at reflecting. For a film 100 nm thick, about the thickness of commercial leaf, the model transmits 0.55% of daylight with a peak at 496 nm. Normalised for brightness, that light is a blue green at a hue of 183.7 degrees, close to #00C4AF. Faraday described the green of light through gold leaf in 1857. Hold a sheet up to a lamp and the yellowest metal there is turns teal.

Light transmitted through 100 nm gold leaf
#00C4AF · 0.55% of the light, peak 496 nm, shown at a fixed brightness

Which gold should you use?

If a brief or a stylesheet says gold and nothing else, the answer is #FFD700, RGB 255, 215, 0. That is the standard, and it is a bright orange-leaning yellow.

If you want a flat colour that people will read as the metal, use goldenrod #DAA520 or metallic gold #D4AF37. Both have the metal’s hue and the depth of gold reflecting gold. If you want it to look like metal, use a gradient that runs from a cream highlight near #FFDEA3 through #FFC25E to a brown near #7D5C29, and keep the hue between 75 and 82 degrees all the way down. The pale highlight is the part most people leave out, and it is the only part that is the colour of gold in the strict sense.

Golds are a good test of colour memory because the family is narrow in hue and wide in lightness and chroma, so most of what you have to remember is depth. The main colour memory game shows a colour, hides it and asks you to rebuild it, and the Hex variant asks for the code itself, if you want to find out how close your idea of #FFD700 really is. Name That Color does the reverse and puts names to swatches.

For related reading, see how to read hex color codes, iridescent colors for another surface whose colour depends on how you look at it, and metamerism for why two different spectra can share one hex.