Researchers from Loyola University in Chicago found that students were more likely to gain weight if they had friends who were heavier than they were.
However, they were more likely to either slim down, or at least gain weight at a slower pace, if their friends were leaner than they were.
Dr David Shoham, who led the study, said their findings, published in the journal PLoS ONE, could help them develop better interventions to treat obesity in teenagers.
"We should not be treating adolescents in isolation," he was quoted as saying by the Daily Mail.
The study was designed to find out the reason why obesity and related behaviours appear to cluster in social networks. The researchers wanted to know whether it is because friends influence one another's behaviour or is it that teenagers befriend people who look similar to themselves.
They examined data from two large high schools -- one refered to as Jefferson High located in a rural area and has mostly white students and another mentioned as Sunshine High was an ethnically diverse urban school.
More From This Section
Students were surveyed during the 1994-95 school year and surveyed again the following school year. Researchers examined data from 624 students at Jefferson High and 1,151 students at Sunshine High. Their body mass index was calculated from their height and weight. A BMI over 25 is considered overweight and a BMI over 30 is considered obese.
The researchers found that even after controlling for this friend-selecting process, there still was a significant link between obesity and a student's circle of friends.
If a borderline overweight student at Jefferson School had lean friends there was a 40 per cent chance the student's BMI would drop in the future. However, if they had obese friends there was a 15 per cent chance they would slim down.
The findings show that social influence "tends to operate more in detrimental directions, especially for BMI", said the authors.
"Effective interventions will be necessary to overcome these barriers, requiring that social networks be considered rather than ignored," Dr Shoham said.
He noted that the study relied on self-reported data and was collected over a decade ago. Nevertheless, the results raised important questions about peer influence, he said.
"Our results support the operation of both homophily and influence," he added.