A BI-DIRECTIONAL NEURAL NETWORK MODEL FOR 3-D OBJECT MOTION PREDICTION IS PRESENTED. MOTION DETECTION AND PREDICTION REQUIRES VERY FAST COMPUTATION SPEED AND HIGH ACCURACY. ARTIFICIAL NEURAL NETWORK, A MASSIVELY PARALLEL AND DISTRIBUTED COMPUTATION ARCHITECTURE, IS VERY SUITABLE FOR THIS SPEED-CRITICAL REAL-TIME COMPUTER APPLICATION. DERIVED FROM SIMPLE NEURAL NETWORK MODELS, A BI-DIRECTIONAL DYNAMIC ASSOCIATIVE NEURAL NETWORK (BDANN) CAN PREDICT OBJECT MOTIONS ADEQUATELY FOR REAL-TIME APPLICATION. THE NETWORK APPLIES A RETROSPECTIVE AUTOREGRESSION MOVING AVERAGE (RARMA) COMPUTATION SCHEME FOR CONTINUAL UP-DATING OF NETWORK PARAMETERS. SIMULATION RESULTS SHOW THE EFFICIENCY AND ACCURACY OF THE APPROACH.