What does that mean? It means that you can estimate where the subject was looking at directly from the functional BOLD image data of your MR scanner! This even works to some degree on old data sets (quality of estimation depends on actual eye movement).