In the field of transportation, telematics is used to obtain vehicle information using Global Positioning System (GPS) technology which is integrated with sensors so that vehicle information can be monitored. One of them is fuel monitoring. The fuel sensor has good accuracy in stationary conditions, but the tability of the data is disturbed when the vehicle is running on an uneven road and causes the tank to shake. This study discusses a fuel sensor noise reduction system using a Kalman filter to overcome the problem of data instability due to shocks. This research aims to reduce noise so that the filter results are closer to the actual result. Filtering is done by changing the process error covariance (Q) and measurement error (R) in the Kalman filter. The fuel sensor noise is simulated using a simulator tank driven by an actuator that can tilt towards the x-axis and the y-axis to resemble the behavior of a vehicle. The fuel level data from the sensor readings are sent by GPS via the cellular network to a server which is then filtered using a web application. From the test results obtained the best filter with (Q) equals 0.1^3 and (R) equals 0.1^3. The average error of the best filter results is 4.73% where this value is 1.92% smaller than the average error of sensor data before filtering, which is 6.65%. Therefore, this proves that the system can reduce noise that occurs in the fuel sensor with the Kalman filter.