This study presents an algorithm for predicting the missing markers during optical motion capture in gaits of children with cerebral palsy. The typical prediction algorithms were for post-processing using interpolation, but this method can predict in real-time through extrapolation calculation. In the previous studies about real-time marker position prediction, they were used for healthy participants who can provide various attached markers or a numerous captured gait data. Alternatively this method uses only the previous positions of each attached marker so that can be used for children with cerebral palsy, who have abnormal gait postures and relatively few attached markers. The proposed algorithm can predict the position of the markers in case of not only the occlusion but also the occurrence of measurement noise, and can be tuned the appropriate parameters according to the gait speed of participants and the performance of the optical motion capture system.,