Reading Faces; Our definition of what's normal changes over time.

Byline: Temma Ehrenfeld

Back in the 1870s, Charles Darwin's cousin Francis Galton wanted to define the face of a criminal. He assembled photographs of men convicted of heinous crimes and made a composite by lining them up on a single photographic plate. The surprise: everybody liked the villain, including Galton himself. He reasoned that the villainous irregularities he supposed belonged to criminal faces had disappeared in the averaging process. In the next century, scientists began to show reliably that faces combined digitally on computers were likable--more so than the individual faces from which they were composed. Although people clearly admire the long legs of Brazilian model Ana Hickmann or Dolly Parton's breasts, in general humans like averages.

Last week researchers confirmed that humans judge real faces by their differences or similarities from a norm. But they also found that the norm can change quickly. When researchers showed 164 people sets of 100 computer-generated faces representing a slow transition from male to female--and from Japanese to Caucasian--it turned out that the test subjects' idea of what constituted an "average" face shifted depending on the first face they saw. When they were flashed a supermasculine face first, more faces on ...