Billionaire art collector R. Crosby Kemper wants his money back.

After meeting Thursday in Kansas City, Mo., with a delegation of experts from the National Gallery of Art and the Georgia O'Keeffe Foundation, the banking tycoon has become convinced that the 28 "early Georgia O'Keeffes" he purchased from art dealer Gerald Peters in 1993 for $5 million are not the work of O'Keeffe. Peters has said he is willing to buy back the works.

Known as "The Canyon Suite," the paintings had been a featured attraction at Kansas City's Kemper Museum of Contemporary Art since it opened in 1994. They were removed from view last month after Kemper was informed by mail that the works had not been included in the new O'Keeffe catalogue raisonne listing all of the artist's accepted works. The 28 watercolors were among 250 works reputedly painted by O'Keeffe that were rejected by four experts representing the National Gallery and the O'Keeffe Foundation.

Kemper initially reacted with anger to the exclusion of his paintings, accusing the National Gallery of spitefulness over his failure to place the works there, which NGA officials had urged him to do in the early '90s. But after sitting down with the four experts, he changed his mind. "In fact," Kemper Museum Director Dan Keegan said, "we're convinced at this point that they are not the works of Georgia O'Keeffe."

Kemper directed all questions to Keegan.

The four experts--National Gallery curator Ruth Fine, paper conservator Judith C. Walsh, catalogue raisonne author Barbara Buhler Lynes and O'Keeffe Foundation President Emeritus Elizabeth Glassman--focused their argument primarily but not exclusively on the paper on which the watercolors were painted. The style and provenance also raised major questions.

Speaking of the 28 works, Keegan said, "We think there are some wonderful ones here, but none of the papers are right. Judy Walsh discusses specifically in the catalogue raisonne that O'Keeffe worked with relatively few types of paper throughout her career, and the problem with 'Canyon Suite' is that none of those papers have been used in the creation of the works. One would expect to find some of the works on that paper."

He added, "This information about the papers was not available back in 1993 when Mr. Kemper was looking at these works. The catalogue project was just getting underway."

Keegan described Thursday's meeting as "very cordial." Asked about any lingering resentments against the National Gallery, he said, "We just decided to set all of that aside. They convinced us that their investigation was not complete in '93 and that the information they have now was not available then."

Asked whether he believed the works were fakes, he said, "Fake is a difficult word, because we don't know how these works came to be, and we don't know the intent. My definition of 'fake' is that someone creates a work of art to look like the work of another artist. And do we have that in this case? I don't know. Because the way these works came to be is not clear to us. We're going with the facts, not the conjecture. And the facts presented to us prove that these could not be the works of Georgia O'Keeffe."

As for dealer Peters, Keegan said, "We've had discussions with Peters all along. We felt like we needed to keep him informed as a courtesy to him, since he has agreed to take [the paintings] back. He knew the meeting was taking place, and I believe he now knows we are asking for our money back. In the middle of the meeting, Mr. Kemper excused himself and made a few phone calls."

Reached yesterday at his gallery in Santa Fe, N.M., Peters sounded mostly on the same page. "I'm going to buy him out and proceed on a technical basis. Or I should say I'm probably going to buy him out--we don't have a done deal, but that's the direction it's heading. I want to have my own people do a paper analysis. . . . At this stage, it's not appropriate to talk about it."

Peters denied that he had spoken with Kemper.

"I still know no more than when we talked," he said. "I want to have real proof [that] the dates of the paper and the date of the paintings are what they claim. . . . Nobody's given me any evidence yet--I still have no evidence whatsoever. All I have is third-hand comment."

In the end, it appears that only the amount of the settlement remains to be negotiated, given Peters's offer to take the paintings back. Kemper said this month that he acquired them in a complex deal, handing over $3.6 million in cash as well as property in Colorado Springs, Colo., that was then valued at $1.4 million. Peters has said that the property was worth less than that when he later sold it.

While Kemper and his museum have become convinced that the works were not painted by O'Keeffe, the question of who did produce them remains a mystery.

After the meeting Thursday, Keegan said, "I'm going to go home and weep. We're exhausted and we're sad. But it's good that we now have this information. Will we ever know for sure who made these watercolors? Or know the real story? I'm not sure we ever will."

CAPTION: "Gray Abstraction (Train in the Desert)," one of 28 works experts say were wrongly attributed to O'Keeffe.