Yeah, I wish I had a link to the original test, still. The testing protocol was rather clever.

And yes, it was fully and truly double-blind. They randomized the order of the tracks burned onto the CD (within each group), then coded the actual track order onto a spreadsheet. They made a few sets of CDs, then assigned each one a random serial number, and randomized which end-listener got which CD. Then the end-listeners got to listen to each CD on their own favorite equipment and report their findings back along with the CD serial number.

Then a simple spreadsheet program correlated the results based on serial number.