Staff and students had lessons on good nutrition. The message was reinforced in other subjects: food labels were used to help teach fractions. And parents were also enlisted: a fundraiser successfully substituted fruit salad for baked goods, said another of the researchers, Sandy Sherman, the Food Trust's director of nutrition education.

She said the children were also urged to exercise at activity stations during recess. They were measured and weighed periodically and surveyed about food and exercise.

After two years, besides fewer new overweight children, the overall number of overweight students at the five schools dropped about 10 percent to 15 percent. At the no-change schools, the number of overweight children rose a quarter to 20 percent.

There was no difference between school groups in new obese students (6 percent) or the overall number of obese (about 25 percent). Obese children probably benefit more from individually targeted efforts, Foster said.

Mike Prelip of the UCLA School of Public Health said the study design was rigorous and the results interesting.

"One intervention usually won't work for everyone," he said. "That's why it's good to have multiple approaches."

The researchers said their findings suggest such programs should start earlier, include gym classes and food sources outside school. Temple University and the Food Trust are now working with corner stores to get them to offer more cutup fruits and vegetables, water, and single-serving snacks. Sherman said they found that local students spend about $2 a day buying snacks that average 600 calories.

Last week, a group of fourth-graders at Fairhill School graded their corner stores during a nutrition class. Crystal Hernandez, 10, said her local store got the top green score while 9-year-old Gabrielle Hudnell's store got a cautionary yellow.

"My mom buys healthy foods now," said Gabrielle. "We have three packs of strawberries, grapes and bananas."

Their favorite snack? Well, potato chips. "I get the little bag," Crystal was quick to add.