Excerpt from the article "Kramer??s Conundrum" by Michael Shermer,
published in the November 29th, 2006 newsletter of Skeptic magazine:

Consciously and publicly, most of us are colorblind. And most of us, most of the time, under most conditions, believe and act on that cultural requisite. You??d have to be insane to publicly utter racist remarks in today??s society ? or temporarily insane, which both science and the law recognize as being sometimes triggered by anger. And alcohol ?? recall Mel Gibson??s drunken eruption about Jews, or the college Frat boys slurring alcohol-induced insanities about blacks and slavery in Sacha Baron Cohen??s film Borat.
The insidiousness of racism is due to the fact that it arises out of the deep recesses of our unconscious. We may be utterly unaware of it, yet it lurks there ready to erupt under certain circumstances. How can we know this? Even without anger and alcohol, Harvard scientists have found a method in an instrument called the Implicit Association Test (IAT), which asks subjects to pair words and concepts. The more closely associated the words and concepts are, the quicker the response to them will be in the key-pressing sorting task (try it yourself at https://implicit.harvard.edu/implicit/).

The race test firsts asks you to sort black and white faces into one of two categories: European American and African American. Easy. Next you are asked to sort a list of words (Joy, Terrible, Love, Agony, Peace, Horrible, Wonderful, Nasty, Pleasure, Evil, Glorious, Awful, Laughter, Failure, Happy, Hurt) into one of two categories: Good and Bad. No problem.

The next task is a little more complicated. The words and black and white faces appear on the screen one at a time, and you sort them into one of these categories: African American/Good or European American/Bad. Again you match the words with the concepts of good or bad, and faces with national origin. So the word ??joy? would go into the first category and a white face would go into the second category. This sorting goes noticeably slower, but you might expect that since the combined categories are more cognitively complex.

Unfortunately, the final sorting task puts the lie to that rationalization: This time you sort the words and faces into the categories European American/Good or African American/Bad. Tellingly (and distressingly) this sort goes much faster than the previous sort. I was much quicker to associate words like ??joy,? ??love,? and ??pleasure? with European American/Good than I did with African American/Good.

I consider myself about as socially liberal as you can get (I??m a libertarian), and yet on a scale that includes ??slight,? ??moderate,? and ??strong,? the program concluded: ??Your data suggest a strong automatic preference for European American compared to African American.? What? ??The interpretation is described as ??automatic preference for European American?? if you responded faster when European American faces and Good words were classified with the same key than when African American faces and Good words were classified with the same key.?

But I??m not a racist. How can this be? It turns out that this subconscious association of good with European Americans is true for everyone, even African Americans, no matter how color blind we all claim to be. Such is the power of culture.