Abstract: Recently, more and more human-computer interaction applications are used cross multiple devices, but existed eye tracking approaches only support single device application, and do not meet the requirement of cross-device interaction. An eye tracking approach oriented to cross-device interaction was proposed. We combined pupil area detection and pupil center detection to reduce the difficulty of pupil image recognition. We predicted position of the missing Purkinje image during the recognition process, and then computed the vector of pupil-Purkinje image. On the other hand, we utilized edge detection algorithm to solve device screen recognition problem, and then created the list of device screen coordinates. Later, we used the coordinates from the list to compute the shapes and sizes of the device screens and identified different devices accordingly. Then we computed the fixation coordinate based on vector of pupil-Purkinje image, and improved the accuracy of fixation coordinate computation with error compensation when there were errors caused by head movement. Finally, a head mounted eye tracker system named as Multi Gaze was developed, and the test results indicated it had high accuracy of fixation computation for cross-device interaction.