Café Wall Illusion

September 2025

Interactive Café Wall illusion demo with adjustable grid, mortar and offset parameters to explore illusion strength.

The Café Wall illusion

What you see

In the Café Wall illusion, rows of offset dark and light tiles are separated by thin “mortar” bands. Although the mortar bands are perfectly parallel and straight, they appear to tilt in alternating directions (a wedge-like distortion). The effect is strongest with high tile contrast, thin, mid-grey mortar, and a half-tile row offset. (Gregory & Heard, 1979; Bach, n.d.; Illusions Index, n.d.). richardgregory.org michaelbach.de The Illusions Index

Why it’s called the “Café Wall”

The illusion takes its name from a tiled wall on St Michael’s Hill in Bristol. Richard L. Gregory, a University of Bristol psychologist, reported the effect after a lab colleague (Steve Simpson) noticed it on the café frontage; Gregory and Priscilla Heard later published the scientific account. A blue plaque now marks the site. (Gregory & Heard, 1979; Gregory, n.d.; University of Bristol, 2017). richardgregory.org University of Bristol

Earlier relatives exist: Hugo Münsterberg discussed a related “shifted chequerboard” in 1894, and A. H. Pierce described the “kindergarten patterns” in 1898. The Bristol observation re-popularised the effect and gave it its modern name. (Pierce, 1898). Zenodo Wikipedia

How to make it pop (and how to kill it)

Tile contrast: Larger dark–light differences strengthen the tilt; low contrast weakens it. (Gregory & Heard, 1979). richardgregory.org
Mortar width: Thin mortar (a few per cent of tile size) is most effective; very thick or vanishingly thin mortar reduces the effect. (Gregory & Heard, 1979). richardgregory.org
Mortar luminance (“mid-grey” rule): Mortar near the mid-grey between the tile luminances maximises the tilt; mortar close to the black or white tile luminance collapses it. (Kitaoka, Pinna, & Brelstaff, 2004; Wolfram MathWorld, n.d.). psy.ritsumei.ac.jp MathWorld
Row offset: Around ½-tile per alternate row gives a strong illusion; 0 offset (aligned columns) largely removes it. (Gregory & Heard, 1979). richardgregory.org
Regularity: Excess row jitter, heavy corner rounding, or strong luminance imbalance can disrupt the global tilt. (Lulich & Stevens, 1989). SpringerLink

A short history

1979 — Border locking: Gregory & Heard proposed border locking to explain the effect: edges from neighbouring high-contrast tiles “lock” across the mid-grey mortar, displacing apparent border positions and creating a wedge pattern along the mortar lines. (Gregory & Heard, 1979). richardgregory.org
1980s–1990s — Filtering accounts: Work on spatial filtering showed that standard centre–surround (DoG) and elongated (oriented) filters produce local “twisted-cord” diagonals in the mortar, which integrate into the global tilt. (Morgan & Moulden, 1986; Lulich & Stevens, 1989). ScienceDirect SpringerLink
2000s — Contrast polarity & constraints: Contrast polarity determines the direction (and presence) of tilt: alternating polarity units drive opposite local tilts that can cancel. (Kitaoka, Pinna, & Brelstaff, 2004). psy.ritsumei.ac.jp

What’s going on under the bonnet?

1) Early spatial filtering + “twisted-cord” cues

Applying standard retinal/cortical filters (e.g., difference-of-Gaussians, oriented receptive fields) to the pattern yields alternating oblique energy in the mortar (a Fraser twisted-cord texture). Locally tilted segments accumulate into a global impression of slanting bands. (Morgan & Moulden, 1986; Lulich & Stevens, 1989). ScienceDirect SpringerLink

2) Border locking across mid-grey mortar

Where mortar sits between the tile luminances, illusory shifts of border location from neighbouring tiles lock across the gap, producing the wedge geometry that your visual system mistakenly reads as tilt. (Gregory & Heard, 1979). richardgregory.org

3) Contrast polarity as a direction selector

If the elementary “units” alternate in contrast polarity (e.g., light body with dark “tail” vs dark body with light “tail”), they drive opposite tilts; arranging polarities appropriately reverses or cancels the illusion. (Kitaoka et al., 2004). psy.ritsumei.ac.jp

Take-home: No exotic circuitry is required; ordinary spatial filters + mid-level edge interactions are sufficient, with contrast polarity and mortar geometry deciding strength and direction. (See review: Westheimer, 2008). CORE

Parameter sensitivities (practical recipes)

Max the tilt:
High contrast tiles (e.g., ~10% vs ~90% lightness), mortar width ≈ 2–6% of tile size, mortar near the mid-grey of the two tiles, row offset ≈ 0.5, minimal jitter. (Gregory & Heard, 1979; Kitaoka et al., 2004). richardgregory.org psy.ritsumei.ac.jp
Make it vanish:
Set offset = 0; or mortar = 0; or set mortar very dark/light (near tile values); or crush contrast; or add row jitter/irregular spacing. (Gregory & Heard, 1979; MathWorld, n.d.). richardgregory.org MathWorld

Why it still matters

The Café Wall is a compact showcase of how early spatial filtering and mid-level organisation (edge interactions, polarity grouping) construct perceived orientation and geometry—sometimes incorrectly. It continues to guide computational models of early vision and to test ideas about contrast polarity and context-dependent tilt. (Lulich & Stevens, 1989; Westheimer, 2008). SpringerLink CORE