Search for neon colors and you get the same answer from every result on the page. Neon colors are very bright versions of ordinary colors, they appear to glow, and here is a list of hex codes. Neon green is #39FF14. Neon pink is #FF10F0. Neon blue is #1F51FF. Copy them into your design and you have neon.
Nobody checks the codes. We did, and the result is blunt enough to say in one line: none of them are neon. They are not unusually bright, they are not unusually saturated, and the thing that makes a real neon surface look the way it does is something a display is physically incapable of doing. The whole category is a set of coordinates borrowed from the edge of a box.
The neon hex codes are just the corners of the box
Every color a normal screen can show lives inside a cube. Three channels, each running from 0 to 255. There is a surface to that cube, and past the surface there is nothing at all. We took the twelve hex codes that circulate most often on neon palette pages and asked a simple question of each one: how far is it from that surface?
| Name | Hex | L* | C* | To the edge | Light back |
|---|---|---|---|---|---|
| Neon green | #39FF14 | 88.2 | 116.0 | 0 | 72.4% |
| Neon lime | #CCFF00 | 93.6 | 99.5 | 0 | 84.4% |
| Neon yellow | #FFFF33 | 97.2 | 90.0 | 0 | 93.0% |
| Neon cyan | #0FF0FC | 86.7 | 46.2 | 3 | 69.5% |
| Electric blue | #7DF9FF | 91.4 | 35.7 | 0 | 79.3% |
| Neon coral | #FF6EC7 | 66.9 | 67.3 | 0 | 36.5% |
| Neon orange | #FF5F1F | 61.3 | 86.3 | 0 | 29.5% |
| Hot magenta | #FF00FF | 60.3 | 115.6 | 0 | 28.5% |
| Neon pink | #FF10F0 | 59.8 | 109.1 | 0 | 27.9% |
| Neon red | #FF3131 | 55.8 | 90.1 | 0 | 23.7% |
| Neon purple | #BC13FE | 49.9 | 116.8 | 1 | 18.3% |
| Neon blue | #1F51FF | 43.4 | 102.1 | 0 | 13.4% |
Ten of the twelve sit exactly on a face. At least one channel is pinned at 0 or at 255. The average distance from the surface across the whole set is 0.3 out of 255, and the two that miss are missing by one and three units, which is nothing. The median HSV saturation is 90 percent.
That is the entire definition, once you strip the adjectives off it. A neon hex code is a color that has been pushed until the screen stops. It carries no information about glow, about brightness, about fluorescence or about neon gas. It says only this: I went as far as this device goes. Which is why the lists never agree with each other and never need to. Any point on the cube surface qualifies.
They are darker than the page they sit on
Look at the last column of that table again. It is the luminance factor: how much light the color sends back, with a sheet of perfect white set at 100 percent.
The average across the twelve so-called neon colors is 48 percent. Neon blue returns 13 percent. Neon purple returns 18. Neon red returns 24. Even neon green, the brightest thing most people picture when they hear the word, returns 72 percent, which means it is noticeably darker than the white background of the page you are reading it on.
This is not a limitation of any particular monitor. It is arithmetic. We swept the whole sRGB cube at a 129 step interval, which is a little over two million colors, and the share of them with a luminance factor at or above white is zero. Not a small number. Zero. White is all three channels at maximum, and there is no fourth thing to add. Every color a screen shows is a fraction of its own white.
Now hold a yellow highlighter against a sheet of printer paper in daylight. The ink is obviously lighter than the paper. That comparison is the thing the hex codes are pretending to capture, and it is the one comparison they cannot survive.
Ordinary matte paint beats them
We expected the screen to at least win on saturation. It does not.
There is a hard ceiling on how saturated a surface can be, worked out by David MacAdam in 1935 and known since as the optimal colour stimuli, or the object colour solid. The argument is short. A surface that only reflects can never send back more than it receives at any wavelength, so its reflectance is somewhere between 0 and 1 everywhere. The most saturated thing you can build under that rule is a reflectance that is exactly 1 across one stretch of the spectrum and exactly 0 outside it. Enumerate every such stretch and you have the boundary of everything an ordinary surface is allowed to be.
We enumerated 152,576 of them at 1 nanometre resolution under D65 daylight, then did the same for the surface of the sRGB gamut, and compared the two hue by hue.
Filled bar: the most saturated color sRGB can reach at that hue. Grey bar behind it: the most saturated color any ordinary matte surface could be. The number on the right is the first as a share of the second.
Averaged around the circle, sRGB reaches 67.6 percent of the chroma an ordinary matte surface could have. It is worst in the greens and cyans, where it manages 39.4 percent at hue 165 and 39.5 percent at hue 180, and best in the oranges at 92.0 percent. The same weakness shows up when you compare screens against print in RGB against CMYK, for the same underlying reason: three fixed primaries are a poor way to cover a hue circle.
Compared fairly, holding each neon hex at its own lightness and hue and asking what the best possible matte surface could do there, the screen versions reach 82.5 percent of the ceiling. Neon yellow reaches 70.9 percent. Neon lime reaches 75.8. So a plain, dull, non-glowing piece of painted card, with no fluorescence in it at all, can be more saturated than the color your design software calls neon. The word is doing work the numbers do not support.
What actually makes something neon
The real mechanism is simple and it is genuinely impressive, which makes it a shame that the palette pages skip past it in one sentence.
A fluorescent pigment absorbs light at short wavelengths, in the violet and the near ultraviolet where your eye is close to useless, and then re-emits that energy at a longer wavelength where your eye is very sensitive indeed. The energy is not created. It is moved, from a part of the spectrum you barely see into a part you see extremely well.
Colorimetry has a name for the consequence. The quantity that describes how much light a surface returns at each wavelength, relative to a perfect white diffuser, is the spectral radiance factor. For anything that only reflects, it cannot exceed 100 percent. For a fluorescent material you have to split it into a reflected part and a luminescent part, add them, and the total goes straight through the ceiling.
We built one. Nothing exotic: a white base with 90 percent reflectance, a dye that absorbs everything below about 500 nanometres, and a re-emission band centred at 540 nanometres. We set the quantum yield at 0.60, which is deliberately conservative for this class of pigment, and ran the whole thing under D65 through the CIE 1931 observer.
The total radiance factor peaks at 189 percent at 540 nanometres. Integrated across the visible range, the sample returns 124.2 percent of the light a perfect white diffuser would, under the same illumination. In CIELAB it lands at L* 108.7 with chroma 141.6.
That L* is the whole article in one number. The object colour solid stops at L* 100 by definition, because L* 100 is the white. There is no ordinary surface anywhere in it, at any hue, at any chroma, with a lightness above 100. The fluorescent sample is not near the edge of what matte surfaces can do. It is outside the set entirely, and it is outside in the one direction the set has no room in.
Switch the fluorescence off and keep the same pigment, and the color falls to L* 92.5 with chroma 113.1. The glow alone is worth 15.69 CIEDE2000 units, which is 6.8 just noticeable differences. And the nearest color a screen can produce is #C6FF00, sitting 11.20 units away, or 4.9 noticeable differences short. Your monitor is not slightly off. It is not in the running.
We ran two more. A modelled orange-red in the Rocket Red family comes out 27.85 units outside sRGB, 12.1 noticeable differences, which is the largest gap of the three and matches the everyday experience that fluorescent orange safety vests photograph badly. A fluorescent magenta highlighter, though, lands 0.90 units outside, which is under one noticeable difference. That one your screen can essentially show.
We would rather report that than tidy it away. Fluorescence does not automatically break the gamut. It breaks it where the pigment can harvest a wide band of short wavelengths and dump them into the green and yellow region where the eye is most sensitive. A magenta cannot do that, because it has to keep the blue end of its own reflectance intact, so there is much less to harvest. Yellow-green highlighters are the loudest objects in a pencil case for a reason that is specific to yellow-green.
Neon gas is not any of these colors
Separate question, and one that Google asks on its own results page: what color is actual neon?
Neon is element 10. Run a current through it at low pressure and it emits on a set of discrete lines, almost all of them crowded between 585 and 705 nanometres. We summed the commonly listed Ne I lines against the CIE 1931 observer and got a chromaticity of x 0.6538, y 0.3458, with a dominant wavelength of 606 nanometres and an excitation purity of essentially 100 percent, because the long-wavelength end of the spectral locus is nearly straight and a pile of lines along it stays on it.
x 0.6538, y 0.3458 · dominant wavelength 606 nm · the real thing is 0.0094 outside the sRGB triangle in u'v', roughly 2.4 noticeable steps of chromaticity
7.92 CIEDE2000 units away, 3.4 just noticeable differences, and 9.4 degrees off in hue
So the gas itself is an orange-red, it is outside the sRGB triangle, and the swatch labelled neon red on the design blogs is three and a half noticeable steps from it in the wrong direction. Every other color on a neon sign comes from a different gas or from a phosphor coating on the inside of the tube. Argon with a trace of mercury gives the blues. The pinks, greens and yellows are phosphors being excited by ultraviolet, which is the same mechanism as the highlighter, just with the pigment on glass and the excitation supplied by the tube rather than by the sun.
Worth noticing that both of the things genuinely called neon, the gas and the fluorescent pigment, land outside what a display can reach. The hex codes named after them land dead in the middle of it. That is not a coincidence, it is what happens when a name travels without its measurement, the same way indigo drifted into eight different colors and magenta ended up as a color with no wavelength at all.
Why the fake ones still work, and why it runs backwards
None of this stops #39FF14 from looking punchy on a dark background. It does. But the reason is not the one the palette pages give, and when you measure it the effect turns out to run in the opposite direction to where the word neon gets used.
Saturated colors look brighter than their measured luminance says they should. The effect is named after Helmholtz and Kohlrausch, and Nayatani published a predictor for it in 1997 that turns a chromaticity into an apparent-lightness multiplier. A neutral grey scores exactly 1.00 by construction. Here is every hex in the set, against a D65 adapting field.
Neon blue gets a 55 percent apparent-brightness bonus. Neon purple gets 54 percent, hot magenta 47. Neon yellow gets minus 11 percent, and neon lime minus 6. The mean across the set is a 25.8 percent bonus, but the spread is the story.
Which gives two completely different colors both wearing the same word. Neon green and neon yellow are bright because they genuinely carry a lot of luminance, 72 and 93 percent, and they collect no perceptual bonus at all. Neon blue and neon purple carry almost none, 13 and 18 percent, and are only convincing because the eye hands them half again as much apparent brightness for free. Put neon green and neon blue side by side on a white page and the blue collapses, because on white you are judging against a surround and its actual luminance is a seventh of the paper. On black it holds up fine. The palette lists do not mention this, and it is the single most useful thing to know before using any of them.
What we would actually tell you to do
- If you want something to read as neon on screen, put it on a dark surround. Every one of these colors is dimmer than white, so the only way to make one look like it is emitting is to make sure nothing brighter is next to it.
- Pick by luminance factor, not by the label. If the color has to survive on a light background, the top half of that first table works and the bottom half does not.
- Do not expect any of it to print. Fluorescent inks exist and they do reach outside sRGB, but the file you sent is a set of coordinates inside it, and the press will hit them exactly.
- Ignore the specific hex codes. They are arbitrary points on a cube face. #39FF14 has no more claim to being neon green than any of the thousands of other colors sitting on the same face.
The thing worth keeping
The gap between a highlighter and its hex code is a good demonstration of something that is true of color generally. The hex code records where a color sits. It records nothing about how the light got there. A fluorescent pigment and a screen pixel can be pushed toward the same region of color space and still be doing completely unrelated things, one returning 124 percent of the light that fell on it and the other returning 72 percent of its own maximum, and only one of them will ever look like it is switched on.
That is the same lesson as metamerism, arriving from the other side. There, two different spectra matched and fooled the eye. Here, the coordinates match and the eye is not fooled for a second, because the brightness relationship to the surround gives it away immediately. Coordinates are not the color. They are a summary of it, and the summary drops exactly the part that made neon worth naming.
If you want to find out how well you actually hold a color in your head rather than in a swatch panel, the color memory game shows you a color, takes it away, and asks you to rebuild it. The hex mode makes you name the code instead of matching the patch, which is a quick way to find out whether those six characters mean anything to you yet. There are more of them here.
How we measured
CIE 1931 2 degree standard observer at 1 nanometre resolution, CIE standard illuminant D65, CIELAB with a D65 white point, and CIEDE2000 for every distance. The object colour solid came from enumerating every two-transition reflectance across 380 to 780 nanometres, giving 152,576 optimal colour stimuli. The sRGB gamut surface was sampled at 181 steps per axis on each of the six cube faces, 196,566 points. The fluorescent samples are models, not measurements of a specific commercial product, and the parameters are stated above so you can disagree with them. Photon bookkeeping for the re-emission is done in photon counts rather than energy, then converted back. A just noticeable difference is taken as 2.3 CIEDE2000 units throughout, and 0.0040 in u’v’ for chromaticity.