WASHINGTON (AP) _ The government's 1986 automobile crash tests showed poor performances by more a third of the 26 cars tested, including failing grades by two new, low- cost imports and Ford's popular Taurus and Sable.

The annual tests by the National Highway Traffic Safety Administration of a sample of cars has shown a general trend of improved performances over recent years.

Nevertheless, 10 of the 1986 model cars tested had scores that indicate one or both front-seat occupants probably would have been killed in a crash.

The best performers were the Toyota Celica, Chevrolet Nova, Chevrolet Cavalier, Oldsmobile Delta 88 and Buick Century, all of which recorded a head injury rating for both front-seat occupants of less 800.

The highway safety agency generally regards any head injury rating of more than 1,000 as probably being fatal to the occupant.

The cars that exceeded the 1,000 threshold for one or both front seat occupants were: the Isuzu I-Mark, Yugo GV, Plymouth Colt Vista, Saab 9000, American Motors Jeep Comanche, Volkswagen Scirocco, Ford Taurus, Mercury Sable, Hyundai Excel, and Subaru GL.

The government tests, which measure impact on a car's occupants in a 35 mph head-on crash into a barrier, have been controversial for years and the auto industry argues they are misleading because of the potential for variation among tests.

Auto safety advocates, however, have said they represent a clear indicator to car buyers as to the relative protection one may expect as a result of automobile design during a major highway crash.

''The crash test results separate the safer cars from the less safe cars,'' insisted Clarence Ditlow, director of the Center for Auto Safety, who expressed disappointment that more cars did not perform better this year.

The NHTSA tests were conducted over the past eight months with results on the final group of tests released this week. Some tests results were made public previously.

Among the worst performers of the cars tested were the two newly introduced low-cost imports - the Hyundai Excel from South Korea and the Yugo from Yugoslavia. The Yugo had a head injury rating of 1,415 for the driver and 1,318 for the passenger, while the Hyundai Excel just dipped under the 1,000 threshold at 999 for the driver but soared to 2,662 for the passenger.

Both companies in statements played down the signficants of the government test results. Hyundai said its own 35 mph crash tests, conducted after a new seat belt system was installed, showed the Excel performing much better.

Ted Kade, a Hyundai spokesman, said the new seat belt system became available last April and ''provides an increased safety margin ... by locking up faster'' than the system used in the NHTSA test car.

Bill Pryor, president of Yugo America Inc., emphasized that the Yugo ''meets or exceeds'' all federal safety standards including requirements for crash protection which are based on a less severe 30 mph crash test.

He said most cars that undergo the 35 mph crash tests - about one-third more severe than the 30 mph tests - exceed the 1,000 threshold. ''When you start adding the factors of weight, car size and price, clearly the Yugo still represents the best value,'' said Pryor in a telephone interview.

Among the surprises in the 1986 NHTSA tests was the poor showing by Ford's popular Taurus and Mercury Sable models. The cars represent a completely new design and have recorded strong sales since being introduced in December.

While the head injury rating on the passenger side was good for both the Taurus and Sable - 695 and 680 respectively - the rating for the driver jumped to about 1200 for both cars.

''When Ford spends that much money on designing a new car you would think they would design it crashworthy,'' said Ditlow of the Center for Auto Safety.

Ford Motor Co., officials were dismayed about the NHTSA findings, but maintained that extensive attention has been devoted to designing safety into the vehicles.

''We don't understand the results of NHTSA's crash tests on these Ford products,'' said Robert H. Munson, Ford's director of automotive safety, ''Our engineers will study the results of those tests in detail including the crash films, still pictures and data in an effort to determine the cause of the higher hic (head injury criteria) readings.''