Bowes Museum Project

When visitors step into a historic gallery space, what actually catches their eye? Do they study the masterworks, read the wall labels, or quickly scan the room? More importantly, how can curatorial displays help visitors engage more deeply with art?

In a collaborative project between The Bowes Museum and Durham University, we set out to answer these questions by taking eye-tracking technology directly into the museum’s second-floor picture galleries, as well as conducting a controlled lab study to evaluate the impact of interpretive labels.

The Challenge: Historic Spaces vs. Modern Visitors

Visit - The Bowes Museum

Founded in the 19th century by John and Joséphine Bowes, The Bowes Museum houses an internationally renowned collection, including one of the UK’s most significant collections of Spanish art (such as El Greco’s The Tears of St Peter) alongside masterpieces by Canaletto and Goya.

However, displaying a vast historic collection presents real spatial and curatorial challenges. Like many Victorian museums, paintings were traditionally densely hung. In partnership with the Museum’s leadership, who are exploring new ways to present their permanent collections, we wanted to find out:

  1. How do visitors actually navigate and view artworks in the galleries?

  2. How much attention do visitors pay to labels and contextual interpretation?

  3. Can revised, targeted curatorial labels guide gaze and improve understanding?

Two Studies: Moving from the Gallery to the Lab

To capture both ecological validity and experimental precision, our project combined two complementary approaches:

1. The In-Situ Museum Study (Mobile Eye Tracking & AI)

We fitted 124 museum visitors with Tobii Pro Glasses 2 mobile eye trackers, allowing them to freely explore the second-floor picture galleries (Peach, Green, and Blue rooms) while their natural gaze behavior was recorded. Following their visit, participants completed a Gallery Experience survey.

To analyze the massive amount of first-person video data, we introduced an innovative computational method: fine-tuning a YOLOv8 object detection model specifically trained to detect and track paintings, sculptures, and text labels across video frames in real time.

2. The Controlled Lab Study (Fixed Eye Tracking)

In our psychology laboratory, 48 participants viewed 30 paintings from the Bowes collection on an EyeLink 1000 system. We tested four distinct context conditions to examine whether varying label text influenced viewing patterns and aesthetic ratings:

  • Condition A: Title Only (basic attribution)

  • Condition B: Title + Existing Museum Context (historical labels)

  • Condition C: Title + Revised Context (new, focused interpretive text)

  • Condition D: Gallery Wall Information + Title + Revised Context

What We Discovered

  • Most Looks Are Quick, But Deep Engagement Drives Enjoyment: Automated gaze analysis showed that 91.1% of viewing events on paintings were brief glances (<1 second). However, paintings that held attention longer received significantly higher enjoyment ratings.

  • Labels Demand Significant Attention: Visitors spent about 25.4% of their total gaze time reading caption labels, compared to 46.2% looking at paintings. Furthermore, engagement with labels strongly correlated with engagement with the paintings themselves (r = 0.40)—visitors who took time to read were also the ones who looked more deeply at the art.

  • The “Hidden Masterpiece” Problem: A strong correlation emerged between viewing rates and artwork enjoyment. Works like Canaletto’s The Bucintoro Returning to the Molo were viewed and recalled by over 85% of visitors and scored at the top of enjoyment metrics. Conversely, exceptional works such as Goya’s Interior of a Prison and Courbet’s View at Ornans scored exceptionally high in liking among people who saw them, but were missed by three-quarters of visitors due to subdued colour palettes, smaller scale, and hanging locations amidst competing paintings.

  • Captions Actively Steer Gaze: In our lab study, providing contextual labels caused a statistically significant increase in the proportion of fixations falling on specific, relevant Regions of Interest (ROIs) mentioned in the text during a second viewing. Captions do not simply inform visitors—they physically guide where their eyes land on the canvas.

  • Wall Text Needs Careful Handling: Curiously, adding broad introductory wall text (Condition D) somewhat diluted this targeted gaze effect compared to focused object captions alone, suggesting that overly broad narrative context may lead to less directed visual exploration.

Recommendations for Museum Display

Drawing on our quantitative gaze tracking and visitor feedback, we offered practical display strategies to The Bowes Museum:

  • Highlight Low-Salience Masterpieces: Reposition subtle or darker works (like the Goya and Courbet) to more isolated, well-lit eye-level locations so they aren’t overshadowed by brighter, panoramic paintings.

  • Prioritize Clear, Accessible Captions: Since labels consume a quarter of visitor attention and directly drive exploration of key visual details, concise, informative captions are crucial—especially for complex religious or symbolic paintings that general audiences find less immediately accessible.

  • Balance Gallery-Wide Wall Text with Object Labels: Avoid overloading visitors with broad gallery narratives that distract from direct artwork engagement; keep interpretive text tightly linked to visible details.

  • Refine Upper-Level Lighting: Visitors noted difficulties viewing paintings hung higher up on the walls due to glare and low lighting, emphasizing the need for targeted illumination.