The Reagan administration yesterday lashed back at a former highway safety administration employe who has privately published a new version of an auto-buyers guide after the administration decided to discontinue publication.

"I don't think the book should have been published," said Raymond Peck, head of the National Highway Traffic Safety Administration, adding that the Transportation Department's inspector general will investigate the actions of its moonlighting ex-employe, John Gillis.

Gillis, who said he prepared the privately published "Car Book" on his own time, relying primarily on government information available to the public, resigned from his $40,000-a-year job Friday after informing Peck that the book was coming off the presses. He was aided by a second NHTSA employe, Ivy Baer, who said she has no intention of resigning.

"I would have much preferred to have accomplished this task as a member of NHTSA," Gillis said at a press conference yesterday. "It's a very risky venture," he added, saying he invested $30,000 of his personal funds to assure publication when it became clear that Peck was discontinuing government publication. He said he decided to keep his work secret from his superiors before publication because he had "no responsibility to tell anyone."

The object of the war of words is an 80-page paperback, which went on sale this week for $4.95. The Center for Auto Safety, a consumer organization that has criticized Peck's policies sharply, joined with Gillis and Tilden Press publisher Joel Makower in financing the project and is also selling the book.

Peck complained yesterday that 85 percent of the consumer information in the book is old data, already published in the government's January 1981 version of the Car Book--and available to consumers without charge until copies of that edition run out. Much of the new material in Gillis' book, he charged, is either incorrect or "seriously flawed."

The sharpest disagreement between the two concerns the use of results from NHTSA's special crash tests on new cars. Peck, who is reportedly under heavy pressure from auto companies to discontinue the testing, said many of the ratings in Gillis' book for 1982 models are based on tests of 1981 cars.

Gillis said he hired an engineering firm to determine which 1981 model results would apply to 1982 cars because the models hadn't changed structurally in a major way.

And Makower said their approach was the only way to inform consumers about 1982 models before the model year ends, since NHTSA won't complete its 1982 crash tests until next summer. "He Peck has tried to pick at every little thing in an effort to unreasonably attack the book," Gillis complained.

NHTSA is currently conducting a study to determine whether the crash-test program is a valid measure of auto safety. In the test, a single car of each model is propelled into a wall at 35 miles an hour and the crash forces are measured on electronically wired dummies. Peck says he isn't satisfied yet that a single test gives an accurate rating for an entire line of cars. "We simply don't know what you learn when you hit a wall with a car," Peck said, adding that this issue is an example of why he decided to discontinue publication.

In addition to the crash ratings, the new book also provides ratings on child-safety seats, fuel economy of new cars, recall data, information on auto insurance costs for various models, advice for purchasing new cars and handling warranty disputes, and performance ratings of tire brands.

The tire rating list is another bone between Gillis and his former agency. Peck is reviewing the current tire rating regulations, which require tire manufacturers to test tires and grade them according to wear, traction and strength. All but one of the major tire producers have called for repeal of the regulation, claiming a few tests on each brand don't provide valid information.

Underlying the argument is a broader issue of how NHTSA should inform consumers about the auto and tire testing results it obtains from producers. Gillis, a marketing expert who joined NHTSA during the Carter administration to promote consumer information services, maintains that the best method is to summarize all available information in one volume, each year, at minimal cost to the government (about 62 cents per book for printing and mailing). The demand for the initial Car Book--more than 1.7 million have been distributed--supports this approach, he says. Since the government isn't doing it, he is, he said.