Driver assistance for cognition and judgment is effective for reducing traffic accidents. However,environment recognition by the sensor is not perfect in the real world. False and missing alarms may cause dangerous situations or disuse of the system by losing driver's trust. We propose a warning method corresponding to recognition reliability of the sensor in order to suppress reduction of the trust from the standpoint of risk engineering. In this research, evaluation experiments using a driving simulator were conducted for a lane change assist system. Effectiveness of the proposed warning method was verified when using the recognition sensor with some errors.