Water is cyan. Its dominant wavelength is 488.8 nm, and the reason you have never seen that colour in a glass is that a glass is too thin to hold enough of it. Not colourless, not blue, and not a reflection of anything. Water is a genuine colourant, one of the weakest in everyday life, and thickness is the only dial that decides whether it shows up.
I built the colour from the measured absorption spectrum of pure water and rendered it at every thickness from a millimetre to a hundred metres. Seven centimetres of it, about the width of a drinking glass, comes out #FDFFFF, which is 0.97 CIEDE2000 away from plain white. That sits under the threshold where a person can tell two colours apart at all, so the glass is not almost invisible. It is invisible, with a margin. Water first becomes visible at 17.3 cm.
The result I did not expect is what happens to the hue. Take the path from 1 cm to 1 m, a hundredfold increase, and the dominant wavelength moves 0.4 nm while the excitation purity moves by a factor of 92. Water has one colour. Thickness only changes how much of it you are allowed to see.
The question splits in two before you can answer it
Ask what colour water is and there are two different measurements hiding under one question, and they give answers 23 nm apart.
Looking through water is a transmission problem. Light crosses the water once, loses whatever the water absorbs, and arrives at your eye. That is a glass on a table, a filled bath, a snorkeller looking at a fish two metres away. Looking into water is a reflection problem. Light goes down, a tiny fraction gets scattered back up by the water molecules themselves, and crosses the water a second time on its way out. That is standing on a boat looking at the sea.
Those two paths give different colours out of the identical substance. I measured the reflection case in why the ocean is blue, where deep clear water renders #030951 at a dominant wavelength of 465.5 nm. This piece is the transmission case, because that is the water almost everybody is actually asking about. The two answers and the gap between them get their own section further down.
How I measured it
Pope and Fry measured the absorption spectrum of pure water from 380 to 700 nm in 1997 with an integrating cavity, which finally solved the problem that had ruined earlier attempts: in the blue, water absorbs so little that stray light and the walls of the container drown the signal. Their table is the standard reference in ocean optics and the Oregon Medical Laser Center posts it in full.
The model is short and has nothing fitted in it:
- Absorption from Pope and Fry, 380 to 727.5 nm at 2.5 nm steps, converted from inverse centimetres to inverse metres.
- Scattering from Morel’s pure water law, 0.0029 per metre at 500 nm falling as the wavelength to the power of negative 4.32. Over a glass this term is worth less than a thousandth of the total, but it costs nothing to carry.
- The colour of a white surface seen through a path of water, under the CIE 1931 two degree observer and D65, then into sRGB. Dominant wavelength and excitation purity are taken against the D65 white point, and distances are CIEDE2000.
One check before any of it is worth reading. Ask the pipeline where pure water is at its most transparent and it answers 0.00442 per metre at 417.5 nm. The published headline figure from the paper is 0.0044 plus or minus 0.0006 per metre at 418 nm. The model reproduces the result of the source it was built from, so the rest of the numbers can be taken at face value.
A glass of water is genuinely invisible
Here is the everyday ladder. The path is how far light travels through the water before it reaches you, the number after it is the CIEDE2000 distance from plain white, and the last column applies the usual just noticeable difference threshold of 2.3.
Everything down to the drinking glass is under threshold, and the sink is the first thing in a normal house that crosses it. Solving for the exact crossing puts it at 17.3 cm of water. That single number explains the entire folk belief that water has no colour: the containers we drink from are built to fit a hand, and a hand is smaller than 17 cm.
It also explains why the same person will swear water is colourless and then describe a swimming pool as blue without noticing the contradiction. Both are right. The pool is 15 times thicker.
Water has one hue, at every thickness
This is the part I ran three times because I did not believe it. As the path grows, a coloured filter normally shifts its hue as well as deepening, because the surviving band narrows around the transmission peak. Water barely does. Below, the left column is the dominant wavelength, the small figure next to it is how far the hue has moved from the 1 cm case, and the bar is excitation purity.
Dominant wavelength, its shift from the 1 cm case, and excitation purity, for a white surface seen through pure water.
Across the first two decades of thickness, from 1 cm to 1 m, the hue moves 0.4 nm and the purity moves by a factor of 92. In colour terms those two columns are not in the same league. Water is a single fixed hue that arrives at your eye in adjustable quantities, which is closer to how a dye at different concentrations behaves than to how a filter behaves.
The reason is the shape of the absorption curve. Water absorbs 140 times more at 700 nm than at 418 nm, and that ratio is stable across the whole red end rather than being a narrow peak. Removing red is all water does, and doing more of the same thing does not change which colour is left over, only how much of the rest survives. The hue only starts to move once the red is essentially all gone and the absorber begins eating into the green, which happens somewhere past 10 m. By 100 m the hue has finally travelled 23.2 nm and water has stopped being cyan and become blue.
Worth saying plainly, because the ladder above can hide it: the swimming pool and the tropical lagoon are the same colour as the glass on your desk. Nothing about the water changed between them.
Water is cyan and the sea is blue, and both are true
Now the two measurements can be put side by side. These are the same substance with no impurities in either, differing only in whether the light crossed it once on the way to you or went in, turned round, and came back out.
#00CAF6 · dominant 484.9 nm · purity 45.7%
#030951 · dominant 465.5 nm · purity 86.8%
Nineteen nanometres apart in dominant wavelength at that pairing, and 23 nm apart if you compare the thin water most people have actually looked at. The reflection case is darker because it is a fraction of a fraction: only around eight parts in a thousand of the light that goes down ever comes back up, and it pays the absorption bill twice. The transmission case keeps almost all of the light and pays once.
So the arguments people have about this are usually two correct answers colliding. Someone who has swum in a pool says water is a pale cyan. Someone who has been offshore says it is a deep blue that is almost black. Neither of them is describing a different liquid.
How many waters are there
If each step has to be a visible change, the transmission ladder from 1 mm to 200 m contains 45 distinguishable colours. The first boundary is at 17.6 cm, the second at 38 cm, the third at 61 cm. They stay that far apart for a while and then bunch up, because purity rises faster than the log of the path for most of the range.
Two of those 45 land almost exactly on colours that already have names in every browser, which is a pleasant coincidence and also a decent way to picture the numbers:
A metre of pure water is CSS lightcyan to within a hair of the visible threshold, and a hundred metres of it is CSS navy. Two names invented for web pages, sitting on the actual transmission curve of water at two thicknesses a hundredfold apart. If you want the codes themselves, how to read hex colour codes covers what those six digits are doing.
Ice is the same molecule and a weaker colourant
Freeze it and the absorption changes more than you would guess. Using Warren and Brandt’s compiled optical constants for pure ice, the blue end gets dramatically clearer while the red end barely moves.
Ice is 4.8 times more transparent than water in the blue and only 1.2 times more transparent in the red, so as a filter it is four times more selective and at the same time much weaker overall. Those two effects pull in opposite directions, and the weaker one wins: ice needs 1.54 times the path of water to become visible at all, crossing the threshold at 26.7 cm against water’s 17.3 cm.
At matched visibility the two end up remarkably close. Take both to a CIEDE2000 distance of 25 from white and water gets there in 6.35 m at #53DDFA, ice in 8.89 m at #56D7FF. Those two hexes are two in the last digit apart on one channel. You cannot tell ice from water by colour, only by how much of it you need.
Which is why snow is white
Snow is ice, and ice is cyan, and snow is white anyway. The usual explanation is scattering, which is right but skips the step that makes it work. Scattering does not add white light. What it does is cut the distance each photon travels inside the ice before it turns round and leaves, and the colour of ice is a function of exactly that distance.
In fresh snow a photon crosses grains a fraction of a millimetre wide and gets bounced back out after a short cumulative path, which is why the albedo is so high. A centimetre of total ice path renders at 0.09 CIEDE2000 from white, which is around a twenty-fifth of a visible difference. Snow is white because its photons are sent home before the ice has a chance to charge them anything.
Push the path up and the colour comes back on cue. Poke a deep hole in a snowbank and the bottom of it looks blue, and glacier ice, where light travels metres through a solid block with few bubbles to turn it round, is the colour the ladder says it should be. The material never changed. Only the distance did.
Almost no real water is this colour
Everything above is pure water, and pure water is a laboratory object. Natural fresh water carries dissolved organic matter leached out of soil and leaf litter, the stuff that makes a peat stream look like weak tea. It absorbs strongly in the blue and falls off towards the red, so it is the exact opposite of water and it wins easily.
Modelling it the standard way, as an exponential decay away from 440 nm, and putting it in a 1 metre path with the water:
Left column is the dissolved organic absorption at 440 nm in inverse metres. One metre path, transmission.
At 0.05 per metre the water is already a visible step away from pure, and at 0.148 per metre the dominant wavelength leaves the blue and cyan side of the spectrum entirely. Values that low are unremarkable in lakes and rivers, and brown water sits one or two orders of magnitude above them. Water’s blue is the most easily bullied colour I have measured on this site. It takes nearly a fifth of a metre of water to see it at all, and a trace of dissolved leaf to cancel it.
That also settles the swimming pool question, which comes up whenever this topic does. A pool is blue because 2.5 m of very clean water is genuinely blue, helped along by a pale plaster or liner that reflects the light back up for a second pass. A pond of identical depth is green or brown because of what is dissolved in it, not because of the mud at the bottom.
Seeing it for yourself
You need a long thin path of clean water and a white reference next to it. A clear length of pipe or a tall florist’s vase, filled, laid on a sheet of white paper in daylight, and looked at end on, gets you past 20 cm without much effort. Put a second sheet of paper beside the far end so you have white and white-through-water touching each other, since a difference of 3 CIEDE2000 is easy to spot as an edge and nearly impossible to spot in isolation.
What you should not do is trust a photograph of it. Every camera applies an automatic white balance, and a faint uniform cyan cast across the frame is precisely what white balance exists to remove. The effect being measured here is a fraction of a percent of purity, so the camera will delete the answer and hand you back a picture of white paper.
The wider point about weak colours
Water is an unusually clean demonstration of something that applies to every colour judgement. The hue was never in doubt. Pure water at 1 cm and pure water at 1 m differ by less than half a nanometre of dominant wavelength, so anyone arguing about whether water is cyan or blue is arguing about purity while using the vocabulary of hue. That confusion is everywhere once you look for it, and I ran into the same thing measuring flamingo pink, where pink turned out to be a purity coordinate rather than a hue, and iridescence, whose honest hex code is grey.
It is also a reminder of how much of colour perception is comparison rather than absolute judgement. Nobody has ever failed to see the colour of water because their eyes are inadequate. A 1 CIEDE2000 difference is at the edge of detectable when two samples share an edge, and completely gone when they do not. Water hides in plain sight because it never gives you anything to compare itself to.
If you want to find out how fine your own threshold really is, the colour memory game is the honest version of that test, since it takes away the side by side comparison and makes you hold the target in your head instead. The spot the difference mode keeps the comparison and shrinks the gap until you lose it, and hue sort is the one that will tell you whether small hue steps or small purity steps are your weak spot. There is more of the same in the full set of colour games.
For the other half of this question, the one about looking down at the sea rather than through a glass, why is the ocean blue has the reflection model and the finding that 82.9 percent of what you see at the surface is reflected sky anyway. And if the whole idea that a transparent substance has a colour still feels wrong, the same argument applies to a mirror, which is #F1F8F7 and green because of its glass.