He comes on stage purposefully, turns to his glittery piano and, after flipping his white tails, plays a striking series of arpeggios. To delighted applause, he ducks back into the wings.

He is, in fact, a duck -- trained here by Animal Behavior Enterprises Inc., a small organization of biologists and technicians that can control the actions of animals as large as whales and as small as cockroaches with behavior modification.

At the company's noisy 25-acre farm in central Arkansas, parrots roller-skate, racoons play basketball, bunnies run roulette wheels, and chickens do card tricks. "We can train any animal to do anything within its physical limits," says Marian Breland Bailey, who founded Animal Behavior Enterprises with her late husband, Keller Breland, in 1947. Since then she has helped train more than 8,000 animals from almost 200 species -- including turkeys, dolphins, cows, snakes and reindeer.

To make research and animal training economically feasible, the organization uses its trained animals commercially for entertainment -- leasing them to shopping malls and fairs for coin-operated shows and exhibiting them at Animal Wonderland and the I.Q.Zoo, the company's own tourist attractions in Hot Springs.

Animal Behavior's training technique, which is approved by the American Humane Association, strengthens and then controls part of an animal's natural random behavior -- say, a chicken's pecking -- by immediate positive reinforcement. If reinforcement is consistent, the desired behavior is likely to be performed again and with increased frequency, and eventually on command.

To show how foolproof the technique is, Bailey has this reporter train a chicken, the animal used most often by Animal Behavior Enterprises in its trained acts. "If you can train a chicken, you can train anything," says Bailey, a tiny, white-haired woman whom everyone rather aptly calls "Mouse."

I am to teach my common barnyard hen ("If she wasn't here, she'd be stew," says one trainer) to pick a winning poker hand by pecking a card marked with a black dot. I have a hand-held button connected to an automatic feeder; when I press it, there is a dull bang and bits of grain fall to a feel tray. The chicken is first conditioned to run to the feeder and eat at the sound of the bang.

"At the beginning of training, reinforce the animal for even a good try at the desired behavior," instructs Bailey. My chicken struts around a bit before she finally gives the eagle eye to the three cards I'm holding. Bang, I reinforce her for looking in the right place. On her next try, she comes close and takes an inquisitive peck at my hand. Bang, I reinforce the pecking action. She pecks a card, any card. Bang, I again reinforce the correct action.

The hen is thereafter reinforced only for pecking the marked card, until, after two hours, she has become an unerring cardsharp. Mastering the training of this barnyard animal carries a special satisfaction for me. At the Texas State Fair last summer, another chicken from Hot Springs dealt me a humiliating defeat in tic-tac-toe.

Once trained, the animals never "forget" the correct response. "You can take Burt Backquack, the piano-playing duck, away from his piano for two years and he'll still remember what to do," says Bailey. "There just aren't that many pianos in the life of a duck."

Most of the animals are reinforced with food, because hunger is the easiest natural drive to work with. (Animals are never starved; the "reward" food is part of their daily ration.) Higher animals like dogs and horses will respond to social reinforcement -- pats, hugs, and words spoken in a kind tone. Company trainers even used darkness as a reinforcement when they were training a cockroach to pull a lever.

Applying knowledge of different species and their habits -- such as the cockroach's love of darkness -- was the Brelands' own contribution to the successful training technique. They also drew from the work of Ivan Pavlov (who, with his salivating dog, originated the conditioned-reflex theory) and from psychologist B. F. Skinner's studies of conditioned response.

Keller and Marian Breland, both psychologists, began their animal behavior studies with Skinner at the University of Minnesota in the early 1940s. In 1941, they joined him in work on his war research project -- a pigeon-guided missile system.

The idea was to have a pigeon keep a missile on target by pecking rapidly at an image of the target on a crosshair screen. Once trained, the sharpeyed pigeons were almost infallible, even when tipped, gyrated and drugged. Nevertheless, the project was dropped.

"The pigeons did exactly what they were supposed to do," says Robert Bailey, the company's staff biologist and now Marian Bailey's husband. "But lets' just say the military lacked faith."

The pigeon project, however, secured the Brelands' faith in B. F. Skinner's animal-behavior work. "They took my theory, applied it well enough to make a living from it and, in the process, had all sorts of useful accidents which we've all learned from," says Skinner, who is retired from his psychology post at Harvard University. The Brelands started their business by training animals for advertising and built it up by doing animal-behavior experiments for the government, most of which are still classified.

One declassified project, field-tested by the Army for use in Vietnam, was an ambush-detection system in which trained pigeons would fly ahead of a convoy and watch for concealed humans. Each pigeon carried a small transmitter that emitted a steady signal as the bird flew. The pigeon was trained to land if it saw anyone lying, kneeling or hiding off the road; when it landed, the signal stopped and the troops were warned.

Animal Behavior Enterprises, which did about $500,000 in business in 1977, stopped all work for the government last year but is still asked to do extraordinary animal training. The Baileys refused a recent request from a Lion Country Safari executive in California who wanted them to train a bull elephant to charge visitors but stop just short of the touring car. "We felt there would be safer ways to star tle people," Robert Bailey explains. They also turned down a man who, for unknown reasons, asked them to teach his donkey to smile.

In addition to wanting to train untried species, the Baileys would like to train animals already in zoos and nature parks to do, on cue, what they do in nature. "We could, for example, train a raccoon to go to a pond and get a crayfish," explains Marian Bailey, "or we could train a bear to hit at something with all his might. People could see the animal in action instead of just lying in the corner of its cage."

So far, zoos have been wary of the idea, even though the Baileys say the natural-behavior exhibits could be made coin-operated.