Iowa State University psychologist Gary Wells said several studies have shown that facial composites constructed from individual features have a low rate of accuracy, while some newer approaches have better success.

Here are some of the key experiments:

In 2005, researchers at the University of Stirling in Scotland had one group of people prepare several composites of 50 British celebrities. When a second group was asked to identify them, they could name only 22 of the 800 composites correctly -- a 2.8 percent accuracy rate.

In 1997, researchers at the City University of New York found that people were unable to tell the difference between composites of classmates they knew and composites of students at other schools they didn't know.

This year, Dr. Wells and his colleagues had students prepare four composites of a target photo. They then "morphed" the four composites together to create a fifth composite that averaged the features of the other four. Then, a separate group was shown the target photo alongside three other photos and asked to decide which photo went with the various composites. That group was able to match the individually prepared composites to the target photo 35 percent of the time -- only 10 percentage points better than chance -- but were able to match the morphed composite to the target photo 48 percent of the time.