A protein in spinal fluid could be used to predict the risk of developing Alzheimer's disease, according to German researchers.
Patients with high levels of the chemical - soluble amyloid precursor protein beta - were more likely to develop the disease, they found.
Doctors said in the journal Neurology this was more precise than other tests.
Alzheimer's Research UK said early diagnosis was a key goal, and the study represented a potential new lead.
Doctors analysed samples of spinal fluid from 58 patients with mild cognitive impairment, a memory-loss condition which can lead to Alzheimer's.
The patients were followed for three years. Around a third developed Alzheimer's.
Those who developed the illness had, on average, 1,200 nanograms/ml of the protein in the spinal fluid at the start of the study.
Those who did not started with just 932 nanograms/ml.
Beta amyloid proteins have already been implicated in Alzheimer's itself, but not as a "predictor" of the disease.
The researchers said that a combination of soluble amyloid precursor protein beta, defective tau proteins, which are involved in the structure of brain cells, and a patient's age was 80% accurate in predicting the onset of the disease.Early diagnosis crucial
There is no cure for Alzheimer's disease. If a treatment is developed, it is thought that it would need to be delivered early, before any permanent damage was done.
Dr Robert Perneczky, from the Technical University Munich, said: "Being able to identify who will develop Alzheimer's disease very early in the process will be crucial in the future.
"Once we have treatments that could prevent Alzheimer's disease, we could begin to treat very early and hopefully prevent the loss of memory and thinking skills that occurs with this devastating disease."
More than 800,000 people have dementia in the UK, and that figure is expected to rise as populations get older.
Rebecca Wood, chief executive of Alzheimer's Research UK, said: "The ability to diagnose Alzheimer's early is a key goal for doctors and researchers. This small study provides a potential new lead to follow up.
"We will need to see larger trials before we can know how accurate this method could be as a diagnostic test. It will also be important to see how measurements of these proteins compare to those found in healthy people."