Abstract: The basic theory of Airborne Scalar Gravimetry is that two series accelerations are different to yield the acceleration of gravimetry which then can be downloaded to the surface of earth.One acceleration is sensed by INS (inertial navigation system)or gravimeter,the other is deduced from GPS observations.High accuracy velocity is required for the calculation of the Etvs corrections (namely Coriolis acceleration caused by the rotation of earth and the movement of airforce) and the determination of the vertical acceleration.And if we want to get the accuracy at mGal level in airborne gravimetry,the kinematic velocity accuracy at cm/s level is very necessary. Several ways of evaluating velocity of moving-base are available by using GPS.Usually the velocity is determined by solving for the position of the vehicle relative to a base station and subsequently taking one time derivative of the three components,which has been successfully used in airborne gravimetry of Switzerland and others.The alternative method is that it can be determined by using GPS Doppler observations,which has advantage of avoiding resolution ambiguity over the first one but isn't as straight forward as it.Due to the largely unpredictable receiver-clock errors and the imposition of the selective availability degradation,double difference Doppler observations are used to obtain the relative vehicle velocity. In this paper the latter theory is discussed,its accuracy is estimated,and the accuracy requirement for the position of airforce or satellite and velocity of satellite is given.Then two tests are tried to validate the reliability and stability of this method and to evaluate the accuracy of this way. First is static test,in which two Trimbel 4000SST GPS receivers are used and respectively mounted on two stations of which coordinate is known precisely,one is looked as base-station,the other as "moving" station whose true velocity is zero.Surveying continued two hours as kinematic survey mode,then observation files are transferred from binary to RINEX format,and the RINEX files are processed by using VAES software(velocity and acceleration estimation system),therefore the result is achieved.We compare the result with the true value and the statistical result indicates that the accuracy are at mm/s level.Second is kinematic test,in which three Trimbel 4000SST GPS receivers are used.One is mounted on the airforce,the others respectively are set on two different base-stations whose coordinate is known precisely.Surveying continued two hours as kinematic survey mode,then two base-stations observation files are respectively processed together with the airforce observation file by using VAES software,and the two results of airforce velocity are achieved.The two series velocities are compared with each other,and the statistic indicates that the accuracy is the same as the first test. Both the theory and two test results indicate that the accuracy at mm/s level of moving-base velocity can be achieved by using this means discussed in the paper.