I measured 3,000 Halloween images across vintage postcards, picture books, catalog pages, pictures of people's front yards, and the last 27 years of Google Doodles to get to the bottom of it.
▦ Halloween pixel scan: click any picture on this page to watch its purple get counted, pixel by pixel.
When I was a kid, Halloween was orange and black. Full stop. But a quick night drive around my community is all the evidence I need to know that is clearly no longer the case. Now, purple is everywhere! When did this happen, and why did this happen?
To find out, I mined the web for a bunch of assorted Halloween imagery and analyzed each image pixel by pixel to try to figure out when purple staked its claim to our favorite spooky holiday. Here's a top level look at my findings. Don't worry, we'll get closer to each category further down the page.
Each stripe is the colored part of that year's images, purple at the top. Hover or tap a stripe for its numbers and sample images. Hatched areas are years with no data; hover one to see why.
As you can see in the color strips above, we found almost no purple in Halloween-themed postcards from 1900 thru 1939, and although the coverage is spotty, we don't see much purple in catalogs and ads from 1969, until 2020, when that year's catalog pages were a meaningful 15% purple.
We have significantly more coverage across picture books and yard displays. Picture-book covers went from almost no purple in the 1970s to about 5% by 1992, and then ticked up to 5% to 7% after that. By 2018 to 2021, purple-share on Halloween picture books was up to 9%.
Yard displays stayed orange much longer. In dated photos of American Halloween yards, purple was about 1% of the strong color through 2015. Since then it has quadrupled, and in 2023 more than a quarter of the displays photographed had noticeable purple in them.
Each dot is one image, placed at its year and its purple share; filled purple means at least 5% of its color is purple, and the strip at the bottom holds the ones with no purple at all. The line is the five-year average, drawn where a full five years of data exist. Hover to see the picture; click to watch its Halloween pixel scan.
Where the data endsThe Internet Archive's lending library holds just three Halloween picture books published after 2017. Its books are scans of physical copies, so recent titles are rare. The shaded years, 2018–2021, use cover images from Open Library instead. Where both sources cover the same years (2010–2017), they agree within half a point: 5.0% vs 4.9% purple for 2010–14, 6.4% vs 6.9% for 2015–17. After 2021 Open Library lists only 1–8 Halloween picture books a year, too few to measure. Yard photos start in 2000 because they come from Wikimedia Commons, which opened in 2004. Only a handful of display photos there are dated before 2007, and recent years are thinner because uploads lag. The poster average is drawn only where five full years of films exist.
As I was pondering how I could find a consistently published Halloween image that would change year over year, it occurred to me that Google's daily Doodle would likely have an annual Halloween rendition, and they keep good records over there.
Sure enough, there's been one every year since 1999. For its first decade it was the usual blue, red, yellow and green letters dressed up with pumpkins, ghosts and a black cat, and 10 of those 11 doodles have no purple at all. Purple arrives with 2010's witch and bats. Since 2017, 7 of 9 doodles are at least 10% purple, mostly in the night skies and the ghosts.
Purple's share of each doodle's color. For scale, the gray lines are picture books and yard displays (five-year averages), which never pass 10%. Hover for each year's doodle.
Doodles© Google, via the Google Doodles archive (doodles.google).
Five-year rolling average, on a square-root scale so the doodles fit. Photos are measured on their strongly colored pixels only, so compare each line's shape, not its level against the others.
Why lines stopEach point averages the five years centered on it, so a line ends two years before its data does. Picture books are dotted from 2015 on, where the average starts to lean on Open Library covers (2018–2021). They end in 2019 because Open Library lists only a handful of Halloween picture books published after 2021. Catalogs stop after 2006: the archive has only two later ones, from 2013 and 2020, too far apart to average. Yard photos start in 2006 because Commons has few dated display photos from before then. Doodles are one image a year, so their average swings with each one. Breaks mark runs of years with no data.
Purple's share of strong color in display photos, by what the photo mostly shows.
Three measures on one timeline. Searches are yearly Google search interest in the US, shown as a multiple of each term's 2004–2014 average. Spending is the National Retail Federation's forecast of what Americans would spend on Halloween decorations. Hover for each year.
SourcesGoogle Trends, US, 2004–2025, all terms in one request so they share a scale. National Retail Federation and Prosper Insights & Analytics annual Halloween surveys, 2013–2025; before 2013 NRF published per-person spending only. Yard purple is the five-year average from the trend chart above.
So, where does all this information leave us? First, we should be honest about the information we're working with here. Counting pixels on images pulled from the Internet probably isn't the method that provides us the most confidence in our answer. Sample sizes are relatively small, we have timeline gaps, all the good stuff that keeps data pros up at night. That said, the data we do have in evidence here does tell a pretty convincing story about purple cementing itself as a Halloween color alongside the rise of LED lighting and inflatable yard decorations.
That's not the explanation you'll usually find. The most-cited one, from an anthropologist writing for Scientific American, is that marketers reached for purple to make Halloween gentler for little kids. The picture-book wave fits that, but the bigger jump came later, and it came in the yards.
It makes sense to me, especially on the lighting front. Black doesn't make a good color for outdoor lights at night, and monochrome orange gets boring fast, so it needs a dance partner, and purple contrasts with it well. Add in the time series we explored in this post and the timing is hard to ignore. Purple shows up in front yards in the same years that Americans started spending real money decorating outside. Searches for Halloween inflatables are more than 4x what they were before 2015. Searches for Halloween lights have tripled, and spending on Halloween decorations in general has more than doubled. Somewhere in that time, lights became a standard part of the Halloween yard.
Three strings of lights against a night sky. The numbers are color difference (CIELAB ΔE): under about 10 the eye barely separates two colors, and above about 50 they're plainly different.
So my best explanation for how purple became a Halloween color is that it simply came within reach. Once a purple string of lights was as easy to buy as an orange one (or more accurately, once a string of lights could be any color or combination of colors), people put them right next to each other.
Hope you enjoyed this.
Kevin Klein
Postcards: 92 Halloween postcards and cards on Wikimedia Commons, mostly 1905–1920; undated golden-age cards count as 1910. Catalogs and ads: about 840 pages from 22 trade catalogs, retail catalogs and store ads in the Internet Archive, 1969–2020 (Tico, Byron Molds, Hallmark, Oriental Trading, Disney, Johnson Smith, Kurt S. Adler, Dollar Tree, Meijer and others). Picture books: covers of 760 Halloween picture books (80 pages or fewer) in the Internet Archive's lending library, 1971–2017. The library often holds several scans of one book; a second scan with the same title, year and cover layout is dropped (85 of them). For 2018–2021, 136 covers from Open Library: works with a Halloween title or subject, 8–48 pages, in English, with coloring, activity and journal titles left out. Lookalike covers count once, and so does a place-name series from one publisher (Sourcebooks' 2019 "Trick or Treat in [State]" books are 50-odd near-identical covers). Yard photos: about 1,050 photos on Wikimedia Commons in the "Halloween decorations in the United States" categories, dated by when they were taken (September to November). Doodles: Google's 27 Halloween homepage doodles, 1999–2025, from doodles.google, measured like the book covers without the white background. Animated doodles are the average of 8 frames across the loop; 2013's was interactive, so only a small still of it survives. They're shown on their own and aren't part of the yearly stripes or averages.
Movie posters: I started from the 645 films in English Wikipedia's "American films about Halloween," "Halloween horror films" and "Films about witchcraft" categories and judged each one from its Wikipedia summary. I kept 136 Halloween movies (set on or around Halloween, or built on trick-or-treating, costumes, haunted attractions or jack-o'-lanterns) and 30 Halloween-season staples that aren't set at Halloween (Beetlejuice, the Addams Family films, Casper, Coraline, The Craft). The witchcraft-only films (fairy tales, witch trials, European folk horror) are out. Posters are each film's Wikipedia lead image, dated by first release on Wikidata, color posters only, 163 in all.
Each image is shrunk, converted to the OKLCH color space and binned by hue: black, red, orange, brown, yellow, green, teal, blue, purple (hue 285°–345°) and pink. Red ends at hue 31°. In the yard photos, red decorations measure 17°–28° and real pumpkins 32°–40°, so anything past 31° counts as orange (or brown when it's dark). Black is anything darker than lightness 0.12, plus nearly colorless pixels (chroma under 0.045) up to lightness 0.36, because printed black ink scans as dark charcoal rather than true black. Dark pixels with real color, like a deep-violet robe or a midnight-purple sky, keep their hue. Pale and low-color pixels (white, cream card stock, gray) are left out. For scans, the page background is detected from a ring just inside the edge and removed, so a purple page backdrop doesn't count as purple product. A zero-shot image model (CLIP) drops catalog pages that aren't Halloween merchandise.
Photos of real displays need different handling. Dusk light tints a white house lavender, so each photo is first white-balanced against its near-neutral surfaces (walls, pavement, sky), and only strongly colored pixels count. Its numbers are on a different scale from the printed sources, and the trend chart is meant for comparing shapes.
The Halloween pixel scan you get by clicking a picture isn't a re-creation. When each image was measured, the label for every pixel it looked at was saved, and the scan replays those labels, so its counts add up to the number used in the charts.
Within a catalog, each page counts in proportion to how much color it has, so a purple header on a text page barely registers. Each catalog counts equally within a year. Each day's photo shoot counts once, so one photographer's 40 frames of a single yard don't swamp a year. The Selling Halloween trade magazine is left out of the catalog row because its own page design is purple. Party City's archived website, also tested, was mostly licensed costumes and is left out too.
Most years before 2007 rest on one to three catalogs, and there's nothing for the 1940s–60s. Commons photos are whatever people chose to upload. That likely tilts toward elaborate displays, and the 2025 count is small (9 shoots). The lending library holds only three Halloween picture books published after 2017, likely because it scans physical copies and new books reach it slowly, so 2018–2021 come from Open Library. Those years rest on 27–48 books each. Resampling the books, the rise from 2010–17 (5.8%) to 2018–21 (9.0%) has a 90% range of about +1 to +6 points, with or without self-published books, so it's probably real but its size is loose. Open Library has too few books after 2021 to use. Postcards thin out after 1920. Their golden age was 1905–1915, and Commons mostly holds cards old enough to be public domain. Yard photos begin in 2000 and are sparse before 2007; Commons opened in 2004. The LED and spending timeline comes from news coverage and industry estimates, not sales of purple lights. Postcards are public domain and the yard photos are freely licensed (credits on hover and on each Commons page). Catalog pages and book covers are copyrighted and appear only as small thumbnails, linked to their Internet Archive pages.