An exam board-backed study found that using new technology to replace traditional tests gives teachers a more negative view of children's abilities.
Martin Johnson of the University of Cambridge Local Examinations Syndicate is today due to tell the Bera conference that although overall differences in performance are small the differences in individual pupils' results can be large and needs further investigation.
The findings are based on the results of 104 10 to 11-year-olds who answered a series of maths questions using both methods.
On-screen GCSEs have been piloted in Northern Ireland, and in England schools are trying out on-screen key stage 3 information and communication technology tests. Supporters say they will increase pupil engagement, save cash and cut teachers' workload.
But the Department for Education and Skills has so far been cautious about reform amid concerns that the results of computer-based testing are not directly comparable with those of traditional tests.
Errors copying information to the screen and a greater reliance on mental calculation, rather than working out answers in writing, contributed to lower scores in the computer-based test, the study found.
Mr Johnson is due to say: "Children may simply have been lazy, preferring to do calculations mentally from the screen, whereas on paper it was natural and easier to show workings on the page."
Pupils may also treat numbers differently on-screen.
"Children may be more comfortable with the idea that numbers set down on paper can be played with and their relationships explored in flexible ways," he will add.
But computer-based tests did not have a wholly negative impact.
The research found that pupils, especially boys, who completed tests on-screen were more likely to attempt questions when they were unsure of the answer.
"One possible reason for this may be that children may link the activity of answering questions on-screen with other activities commonly associated with computers such as games, which may promote a philosophy of have a go and start again," the research found.
"On-line assessment: the impact of mode on student performance" by Martin Johnson and Sylvia Green is available from www.ucles-red.cam.ac.uk