A system and method for recovering lost data in an electronic gyroscope sensor system are disclosed, which use a linear adaptive predictive technique for determining what data was lost by the gyroscope sensor system during a disruptive interval involved. More precisely, a system and method for recovering lost data in a fiber optic gyroscope sensor system are disclosed, which continuously predicts "N" future samples of sensor data and stores the last known good "L" sensor values and the calculated "L" coefficients in a non-volatile memory. In the event that the fiber optic gyroscope sensor system becomes inoperable (e.g., due to a temporary loss of power to the gyroscope or other cause of electromechanical failure), and once the gyroscope sensor system resumes operation (e.g., power is reapplied), the stored "L" coefficients are retrieved from the non-volatile memory, and are used to calculate the data lost by the fiber optic gyroscope sensor system during the inoperative period involved. During normal operation, "N" future samples are predicted. Also, while actual sensor data is available, the actual data is compared with the predicted data, and any resulting differences are applied to an adaptive least mean squares algorithm, which updates the coefficients and corrects prediction error in the linear adaptive predictive filter being used.