Short answer
From sources that say what was observed, where and how — kept in three separate grades: measured, documented and editorially derived. Colourwise's own dataset shows how thin the top grade is: of 57 records, 2 carry a published colour notation, and both describe small samples, not a townscape. A palette sampled from photographs belongs in none of the three grades, because a photograph records a camera's rendering of one light, not a surface.
A record is 'measured' when a source publishes an instrument reading or a colour-order notation for an identified sample: the British Geological Survey's Munsell 5YR 7/1 for a polished specimen of Hailes sandstone, or CIELAB values calculated from the reflectance of six ochre samples at Roussillon. It is 'documented' when a cited source states the material or rule and describes its colour in words, as geological registers and heritage bodies do. It is 'editorially derived' when Colourwise has joined two sources — one for what a place is built of, another for the colour or its cause — as with Aberdeen's granite. The grades are about how the place and the colour are linked, and are tagged separately from how reliable each source is.
A pixel value is the end of a chain: the daylight at that moment, the surface, the camera's white balance and tone curve, and the screen. Change any link and the number changes. Field research shows the surface itself is hard enough. In the Uppsala façade study, the colour perceived from the street always had less blackness than the colour of a sample held against the wall, and shifted between sun and overcast. Daylight measured at one site in Granada ranged widely in colour temperature over two years. A palette picked from holiday photographs therefore averages unknown lights, unknown camera settings and whichever buildings were photogenic, and then reports the result with six-digit precision.
Precision is not accuracy. A HEX value implies a tolerance of well under one per cent on each channel; a description such as 'pale buff' honestly implies a range.
Every record in the dataset states its scope, and only 30 of 57 speak for a whole city or more. Portland stone is recorded for a group of public buildings in London, not for London. The Hailes notation belongs to one specimen from one quarry. Glasgow's cream and red sandstones are recorded for the city centre along a published route. Welsh slate carries a warning that it was exported so widely that a slate roof says little about where it stands. Most misleading 'colours of a city' are true statements about a district, a building type or a period that have been promoted to the whole.
Where a source gives a notation, the dataset reproduces it in the source's system: Munsell for the Edinburgh specimen, CIELAB with its illuminant and observer for the ochres. It does not convert either to RGB or HEX. A conversion needs assumptions the source may not supply — the illuminant a Munsell match was judged under, how a rough stone face relates to a flat chip — and the result invites use as a swatch for a place when it began as a reading of a hand sample. Where this site calculates anything from a source's numbers, such as hue angle from a* and b*, the result is labelled as Colourwise's calculation and the method is given.
8 of the 57 records have no colour at all, because their sources give none. The museum page that documents Lutetian limestone as the stone of Paris describes its age, fossils and quarries but not its colour. UNESCO's inscription for the zinc-roofers' craft says nearly 80 per cent of Paris roofs are zinc and does not say what colour they are. Zanzibar's Stone Town is recorded as plastered and lime-washed, colour unstated. It would be easy to fill those cells from general knowledge. Leaving them empty keeps a clear line between what a source supports and what everyone assumes, and marks where a proper survey is still needed.
The methods exist. The colourist Jean-Philippe Lenclos built regional palettes by collecting samples of render, joinery and roofing on site and recording them systematically. The Uppsala study established 'inherent colour' by holding standard colour samples directly against each wall under the prevailing light and noting the nearest match, then recorded perceived colour separately from a stated distance. A survey meant to be reused should add what those do: the building and elevation, the date and weather, the cleaning state of the surface, the instrument or atlas used, and how many buildings of what type were sampled. A palette with that record behind it is evidence. One without it is an illustration.
| Dimension | Measured | Documented | Editorially derived | Colour stated by source | Total |
|---|---|---|---|---|---|
| Building stone | 1 | 22 | 1 | 23 | 24 |
| Brick and fired clay | 0 | 5 | 0 | 5 | 5 |
| Earth pigment, limewash and earth walls | 1 | 8 | 0 | 8 | 9 |
| Roofing | 0 | 3 | 0 | 2 | 3 |
| Painted tradition | 0 | 9 | 0 | 9 | 9 |
| Heritage regulation | 0 | 7 | 0 | 2 | 7 |
| All records | 2 | 54 | 1 | 49 | 57 |
| Dimension | Strong | Moderate | Limited |
|---|---|---|---|
| Building stone | 20 | 4 | 0 |
| Brick and fired clay | 4 | 1 | 0 |
| Earth pigment, limewash and earth walls | 5 | 2 | 2 |
| Roofing | 3 | 0 | 0 |
| Painted tradition | 2 | 3 | 4 |
| Heritage regulation | 5 | 2 | 0 |
Each statement is labelled by kind — established fact, a standard’s requirement, observed market data, a convention, or Colourwise’s own interpretation or analysis — with the strength of the evidence behind it.
Colourwise analysisModerate evidence
Of the 57 records in Colourwise's place-colour dataset, 2 carry a published colour notation and 8 carry no colour because their sources state none.
Based on: A count of the dataset's provenance tags and of records whose source gives a colour description; see the tables on this page.
Caveat: Counts what Colourwise located and could read, in English; measured surveys may exist that were not found.
Source: Building stones of Edinburgh: introduction; Relationship between Colours of Ochre from Roussillon and Content of Iron-Bearing Minerals
FactStrong evidence· Uppsala region, Sweden
In field observations of 125 façades, the colour perceived from the street always had less blackness than the inherent colour found by holding a colour sample against the wall, and perceived colour differed between sunlit and overcast conditions.
Caveat: One region of Sweden; visual assessment within the Natural Colour System.
FactModerate evidence
Jean-Philippe Lenclos developed a 'geography of colour' method that surveys a region's building colours by sampling materials on site and recording them systematically.
Colourwise interpretationLimited evidence
A colour palette sampled from photographs of a place is not evidence of that place's colours, because each pixel combines an unrecorded illuminant, the camera's processing and the photographer's choice of subject.
Based on: Follows from measured variation in daylight at a single site and from field evidence that perceived façade colour departs from inherent colour; camera white balance and tone rendering add further unrecorded steps. Not itself a tested comparison of photo-derived and surveyed palettes.
Caveat: Calibrated photography with a reference target in frame is a legitimate recording method; the objection is to uncontrolled images.
Source: What colour is the red house? Perceived colour of painted facades (Karin Fridell Anter, doctoral thesis, Trita-ARK 2000:4); Color and spectral analysis of daylight in southern Europe
FactStrong evidence
The British Geological Survey's caption for a Hailes sandstone specimen gives a Munsell notation, 5YR 7/1, without stating the viewing conditions; Colourwise reproduces the notation and does not convert it to another colour system.
Caveat: A single specimen; the editorial choice not to convert is Colourwise's policy.
Reviewed 6 October 2026. Colourwise summarises its sources in its own words and does not reproduce standards text or proprietary colour data. Spotted an error? Tell us.