ORIGINAL: bitflipper
This was a study done in Japan by some major audio manufacturer (like Sony or somebody) where an MRI was used to determine whether or not people could actually hear ultrasonic frequencies. Subjects were asked to A/B music with and without ultrasonics and to subjectively rate them. Subjects could NOT tell the difference, and there was no statistical correlation between their subjective ratings and ultrasonic content. However, the presence of ultrasonic components did cause a measurable effect in the MRI. Their conclusion was that we may be able to subliminally detect them, but that detection does not permeate our consciousness at all.
Actually, bitflipper, I'm surprised you remembered this. However, that isn't exactly how the experiment went down. It was a Japanese study. What was compared was music from LPs and "identical" music from CDs. The actual measuring device was a quantitative EEG (electroencephalograph... recording brain electrical response). Using a frequency analyzer, the experimenters determined that there was significant power in the frequency spectrum from the LPs that extended about 2 octaves from 20K. The CD, of course, produced no energy above 22K. What was interesting about the results, was not that they found that the brain could respond to the inaudable frequencies, but that the response was not immediate. That is, it took a number of minutes for the brain to recognize a difference in the signal and respond to it.
It has been extensively debated in auditory neuroscience circles what this means. One significant question was raised by a group of neuroscientsts (of which I was one). The question we asked was one of perceptual familiarity. In the study (which was conducted in the mid 80s), the subject pool were relatively older (i.e., not students). These people had all "grown up" on listening to music played from LPs. In the test, the LP was compared to CDs for music that they were familiar with (this was to control novelty effect). However, our criticism was in the control of the novelty efect, they were inducing a unique novelty effect, as the main variable (i.e., LP vs CD). There is a lot more than frequency difference that differentiates CDs and LPs. LPs have clicks, pops, and rumble (most people didn't listen to virgin vinyl on an audiophile turntable). The brain could simply have been responding to the absense of this "additional" audible content.
In the mid 90s when I was on sabbitical, i atempted to conduct several experiments that actually possed the familiarity question. By the mid-90s most people of College age had presumably had most of their music listening experiences in the post-CD era. That is, the CD would be the format they were "used" to. Unfortunately, we never came up with a way to test this hypothesis. 192K sampling wasn't very feasible at the time, so there would be no way to preserve the frequency content of an LP while also removing the artifacts of the LP (clicks, pops, rumble, etc). Without being able to remove the audible artifacts, you couldn't do the study correctly, so we passed, and did a study demonstrating that MP3s and CDs did not differ in appreciated quality in a non-A/B test. But, that's another story.
Actual;ly, this study would be really easy to replicate today. Unfortunately I'm retired from science now.