Scientific method / Claim & critique

Ice-crystal claims: why beautiful images need stronger tests

Source year 2006–2017 · Note reviewed 2026-10-12

The 2006 pilot reports that 100 blind judges gave higher beauty ratings to 24 selected treated-condition images than to 16 control images (means 2.87 vs 1.88; one-tailed p=.001). The images came from just four bottles (two per condition), and only drops with a discernible crystal were photographed. The same investigator group’s 2008 triple-blind follow-up used six bottles (two per condition) and 300 drop photographs. Its authors reported a weak three-condition beauty effect (p=.03) and a treated-versus-proximal-control comparison at p=.05, one-tailed; the beauty effect was not significant after adjustment for image interest (p=.20), and the three-condition objective-contrast result was not significant (p=.25). These are reports by the study authors, not independent confirmation.

What this does — and doesn’t — establish

The 2006 image-level ratings do not create more independent treated water sources than the four bottles; the authors themselves noted different treatment/control environments, selective photographing of visible crystals and the need to assess all drops and use more objective measures. The 2008 authors modeled bottles as random effects but used only two bottles per condition. Stokes’s 2009 letter specifically critiques the 2008 treated-versus-proximal comparison: treating the four bottles as the units gives a maximum one-tailed exact rank probability of 1/6 (.167). This letter does not directly analyze the 2006 pilot. The 2017 technical note discusses dependent photographs and too few bottles; its authors also acknowledge their own judge-level tests were invalid and their two-bottle-per-condition trials too small. That small methods study is not a decisive independent null replication. No reviewed study shows that water understands words, carries a message, or reliably changes crystal structure because of thought.

What we can build on

Many photographs and ratings cannot substitute for independent experimental units. A communication claim also needs a defined sender and receiver, a signal, and a controlled functional response; visual resemblance or a human beauty score is not a message. This is a lesson in research design, not biological evidence or a basis for Talk2Nature machine learning.

See the related tools and reading paths →

The next reading task

Keep this as a methods-critical case only. Any new test would need specialist review, preregistration, many independently randomized bottle batches, blinded handling, all-drop measurement, objective outcomes, a frozen analysis and independent replication. Do not use these papers to train or benchmark an animal-communication model.

Go to the sources

1. Ice crystals and distant intention: 2006 double-blind pilot ↗

   Radin, Hayssen, Emoto and Kizu (2006), Explore 2:408–411. Full four-page article reviewed from an author-hosted copy. Four bottles yielded 40 selected crystal images for blind beauty ratings; this is the study authors’ reported pilot result, not evidence of communication.

2. Ice crystals and distant intention: 2008 triple-blind follow-up ↗

   Radin, Lund, Emoto and Kizu (2008), Journal of Scientific Exploration 22:481–493. Full paper reviewed: six bottles, two per condition, and 300 photographs; authors report weak subjective-rating results and a nonsignificant three-condition contrast result. Same investigator group as the 2006 pilot.

3. Ice-crystal statistics: 2009 letter on the 2008 experiment ↗

   Douglas M. Stokes (2009), Journal of Scientific Exploration letters, p. 83. For the 2008 treated-versus-proximal comparison, argues that the four bottles (two per condition), not the repeated dishes/images, are the experimental units and gives a maximum one-tailed exact rank probability of 1/6. This is not a critique of the 2006 pilot.

4. Water-crystal methodology: 2017 technical note ↗

   Kokubo, Nemoto and Kawano (2017), Journal of Parapsychology 81(1):73–86. Full 14-page technical note reviewed. It discusses non-independent crystal photographs and too few bottle-level treatment units, and acknowledges limitations in the authors’ own small tests. It is methodological context, not a decisive independent replication or evidence of communication.

Original editorial note by Talk2Nature contributors. AI-assisted, source-grounded synthesis; not a peer review or an independent replication. Linked works retain their own rights.