Thousands of pupils could end up with the wrong GCSE and A-level grades because of changes in the way they are awarded, an exam board official has warned.
Exam regulator Ofqual is encouraging a much greater emphasis on statistics showing pupils' prior achievement when deciding how exams should be graded, according to Tim Oates, head of research at Cambridge Assessment.
He said this came at the expense of looking at pupils' actual answers and could lead to "grades in the wrong place" for some pupils.
Mr Oates, whose employers run the OCR exam board, claimed that by encouraging this "huge shift", the exams regulator could be breaching its own code of practice.
His concerns centre on the grade boundary decisions that exam boards take to set the number of marks needed in an exam paper to obtain a particular grade.
The process of setting the grade boundaries sees boards take samples of completed, marked scripts and base decisions on two factors: examiners' judgments of pupils' performances and statistical information showing how pupils performed in previous GCSE or national tests. Mr Oates says recent years had seen a "huge shift" towards the latter.
"If you are a young person and you are working really, really hard and you think that what happens on that exam paper really counts, it is quite wrong that the system behind the scenes doesn't actually pay much attention to what you have done," he said.
"This is not an attack on Ofqual - it is just that I believe this is a debate we should have in public."
Recent changes to the structure of GCSEs and A-levels make it harder to ensure that grading is constant over time, and Ofqual says looking at pupils' prior attainment statistics can help maintain standards in such circumstances.
But Mr Oates pointed to two examples of where over-reliance on statistics could cause mistakes.
First, papers used in the sample could be from schools that had only just started teaching a new subject. Basing grade boundaries on prior attainment statistics would take no account of the inexperience of the school and could artificially inflate the grades.
Second, the prior attainment statistics being considered might have been from an anomalous year in which a particularly difficult exam paper or other factors created a "distortion".
"You can't just assume there is an invariant relationship between prior attainment and the grade," Mr Oates said. But he claimed this was exactly what Ofqual had done.
"In some subjects, the Ofqual demand is really pushing us towards 100 per cent reliance on statistics," he said.
"In some of its advice to awarding bodies, it is not impossible that Ofqual has been in breach of its own code of practice, which actually states that the chief examiner should make a professional judgment and should be the person who embodies the standard."
An Ofqual spokesman said its code of practice stated that grade boundaries must be set using professional judgment "informed by the relevant technical and statistical evidence", including information on prior attainment.
He said statistical evidence was "especially important in ensuring standards are maintained when new specifications are introduced, as is the case this summer with the introduction of new A-levels".
The AQA exam board said the use of statistics to support examiners' judgments was "extremely helpful", particularly at a time of structural change in qualifications.