An unapproved type of genetically engineered grass has been found growing in the wild in what scientists say could be the first instance in the United States in which a biotechnology plant has established itself outside a farm.

Ecologists at the Environmental Protection Agency said they had found a small number of the grass plants growing in central Oregon near the site of field tests that took place a few years ago.

The E.P.A. scientists and others said the grass would probably not pose an ecological threat. Still, it could provide fodder for critics who say that agricultural biotechnology cannot be adequately controlled.

“It is a cautionary tale that you have to think about the possibility of plants escaping into populations where there are wild relatives present,” said Jay Reichman, an agency ecologist who is the lead author of a study to be published in the journal Molecular Ecology.

The genetically engineered grass, called creeping bentgrass, is being developed by the Scotts Miracle-Gro Company and Monsanto for use on golf courses. It contains a bacterial gene that makes the grass resistant to the herbicide Roundup, known generically as glyphosate.

The goal is to create a product to allow groundskeepers to spray the herbicide on greens and fairways to kill weeds without hurting the grass.

The Department of Agriculture is evaluating whether to approve the grass. A department spokeswoman said that no timetable had been set for making a decision, but that the new information would be assessed.

One concern often raised by critics of agricultural biotechnology is that genes that make crops resistant to herbicides or pests may escape to wild relatives, creating “superweeds” that would be harder to eradicate.

That is hardly a risk for the main types of genetically engineered crops grown in the United States — soybeans, corn and cotton — because they generally do not have wild, weedy relatives in this country.

But it has been a concern with the genetically engineered grass, which has wild relatives. And, unlike corn or soybeans, grass does not have to be replanted every year.

Some scientists have expressed concern that if the gene escapes, weedy grasses could be harder to control with glyphosate, a widely used herbicide.

Because of those concerns, the Agriculture Department is doing a full environmental impact assessment before making a decision. It will be its first involving a genetically engineered crop.

Two years ago, scientists at the E.P.A. laboratory in Corvallis, Ore., published a paper showing that pollen from a test plot of the grass had spread as far as 13 miles downwind, much farther than many had expected. That made it likely that genetically engineered grass would be found in the wild, though the scientists did not look for that.

In the new study, scientists sampled 20,400 plants up to three miles from the edge of an 11,000-acre zone surrounding the test plots. They found 9, or 0.04 percent, that were genetically engineered, the farthest being 2.4 miles from the control zone border.

The scientists said some of the plants had been created by seeds that had blown off the test plot and others by hybridization of wild grass with pollen from the genetically engineered grass. All were of the same species of grass being developed by Scotts and Monsanto.

A spokesman for Monsanto said that creeping bentgrass lacked the characteristics needed to become a weed and that other herbicides could control Roundup-resistant bentgrass if need be.

Jim King, a spokesman for Scotts, said the company had already admitted that some grass was growing outside the test plots and that the company was working to eradicate it. In field tests, Mr. King said, a windstorm arose when the grass had been cut and was drying in the field, dispersing seeds.

Scotts argues that grass on golf courses, which is kept short, does not pose the same threat of seed dispersal or pollen flow as grass grown to produce seeds.

The company says the nonengineered bentgrass now used on golf courses has not become a weed, and people outside of golf courses do not try to control it by spraying Roundup.

But Norman C. Ellstrand, a professor of genetics at the University of California, Riverside, said that in some parts of the country bentgrass was considered a problem and was controlled. Dr. Ellstrand, an expert on gene flow in plants, said that foreign genes put into crops had escaped into the wild in other cases abroad.

Scientists in Canada have reported an instance in which herbicide resistance appears to have spread by pollination from genetically engineered canola, which is widely grown there, to a wild relative.

In Japan, transgenic canola was found growing near some ports and roadsides. Since the crop is not grown commercially in Japan, scientists hypothesized that imported seeds had escaped during transportation to oil-processing facilities.