Our secret intentions remain concealed until we put them into action - so we believe. Now scientists have found a way to read our minds from patterns of our brain activity. For the first time, researchers have been able to predict how a person has decided to act, simply from viewing a brain scan.

In future, the intention to reply to an e-mail or indicate left could be picked up by brain scanners and turned into the appropriate action. One day, the scanner will be able to read even abstract thoughts from patients' brains, raising fears of invasions of privacy.

We plan hundreds of actions every day. How and where the brain stores these intentions has been revealed by John-Dylan Haynes from the Max Planck Institute for Human Cognitive and Brain Sciences, with colleagues in London and Tokyo.

They were able to "read" participants' intentions from their brain activity by a new combination of functional magnetic resonance imaging (FMRI) and sophisticated computer algorithms. The researchers let subjects freely and covertly choose between two possible tasks - either to add or subtract two numbers. They were then asked to hold in mind their intention for a while until the relevant numbers were presented on a screen.

Researchers could recognise the subjects' intentions with 70% accuracy based on brain activity - even before the participants had seen the numbers and had started to perform the calculation.

Participants made their choice covertly and initially did not know the two numbers they were supposed to add or subtract. Only seconds later, numbers appeared on a screen and they could perform the calculation. This ensured that the intention itself was being read out, rather than brain activity related to performing the calculation or pressing the buttons to indicate the response.

Haynes said: "It has been previously assumed that freely selected plans might be stored in the middle regions of the prefrontal cortex, whereas plans following external instructions could be stored on the surface of the brain. We were able to confirm this theory in our experiments."

The work of Haynes and his colleagues goes far beyond simply confirming previous theories. It has never before been possible to detect from brain activity how a person has decided to act in the future.

The trick by which the invisible is made visible lies in a new method called "multivariate pattern recognition". A computer is programmed to recognise characteristic activation patterns in the brain that typically occur in association with specific thoughts. Once this computer has been "trained" it can be used to predict the decisions of subjects from their brain activity alone.

"The experiments show that intentions are not encoded in single neurons but in a whole spatial pattern of brain activity," says Haynes, in today's edition of the journal Current Biology.

His findings raise hope for improving the lives of paralysed patients. Already the first steps are being made with computer-assisted prosthetic devices and so-called brain computer interfaces. These focus on reading out the movement the patient intends to perform. Previous research has shown that patients can move artificial limbs or computer cursors purely by the power of their mind. But Haynes's findings open up a new perspective.

Meanwhile, "mind reading" technology is becoming big business in the US, where companies have invented "lie detectors" which rely on FMRI scans. Two firms are now offering brain scans for lie detection.