The tests, focusing on car seats, showed nearly two-thirds of models scored poorly in a rear-end shunt - the most common form of accident.
Of 114 models tested by the Thatcham research centre, just 18 were rated ‘good’ and 27 ‘acceptable’. The other 60% were marginal (24%) or poor (36%).
Among models given a ‘poor’ rating were the BMW 3 series, Jaguar X-Type, Honda CRV, Toyota Corolla, Renault Clio, Citroen C2/C3, Mercedes C-Class, Honda Jazz, Land Rover Freelander and Volkswagen Polo.
Twenty-five models automatically failed the tests because their car-seat positioning was such that they did not provide the minimum protection level, and were therefore not tested.
Matthew Avery, the head of the Berkshire-based Thatcham test programme, said: "These results are very disturbing. Some manufacturers - and many motorists - will be shocked by these findings. Our internationally recognised protocols are extremely accurate. Manufacturers will have to make significant changes if they wish to really protect people in a car crash."
The Society of Motor Manufacturers and Traders today reacted with concern over the results. A spokesman says: "The inference that we put unsafe seats in cars is nonsense. The reality is that four out of five whiplash injuries could be prevented simply by adjusting the head restraint properly."
Of particular concern are seats to which Thatcham has assigned a poor or marginal rating on static tests, says the SMMT. These are automatically given a 'poor' rating for overall performance, with no additional dynamic test.
Those that are tested with a sled and 'whiplash crash dummy', are not set up with the head restraint located at an optimum height. Instead Thatcham use a standard mid-height setting which could adversely affect results.
The SMMT believes results could be misleading since they are not supported by real accident data. It called for more research.