n the WSN - WIM system, the sensor fault diagnosis is a critical issue. This paper proposes a sensor fault diagnosis method based on the grey prediction theory, while the gray prediction model was optimized to avoid the theoretical defect of prediction model to the first sensor fault diagnosis caused by false assumptions. Over simulation experiment of WSN-WIM system , achieved an less than 0.3% average error,and 100% diagnostic accuracy. moreover this method has small computation, Good real-time performance, strong ability of self-adaptability, self-diagnostic and so on. It is very suitable for embedded, distributed high-speed dynamic multi-sensor weighing systems.