Star centroiding, locating the star center in a star image frame, is a fundamental process for any star tracker. In this paper, the approximate locations of the stars in successive image frames are predicted using the angular velocity as provided by a rate gyro, then the centroid is updated based upon local image processing. When the rate gyro data are not available, then the angular velocity is estimated using the attitude kinematics equation and successive attitude estimates from the Lost-In-Space Algorithm. Also considered are the special features of noncircular star image shapes associated with optical tagging of starlight and/or image smear. Finally, an approach is presented to implement these ideas with the recently introduced Active Pixel Sensors, allowing dynamic pixel access and selected subarray analog-to-digital conversion of the pixel information is feasible, with logic dictated by most recent image and the instantaneous angular velocity estimate. This novel process that predicts the star image starlight locations is termed predictive centroiding. The problem of the image smear is also treated, in which the relatively high angular velocity of the spacecraft will affect the shape of the star images. These approaches, coupled with active pixel sensors, should enable near-optimal image processing and high frame rates. The paper includes analytical, computational, and night sky experimental results.