Nowadays, it is importantly and urgently needed in Egypt the existence of precise geoid for many applications. For that reason, lot of terrestrial gravity data could be collected as gravity values, gravity anomalies, and GPS leveling undulations. These data might be collected from governmental agencies, petroleum companies, and research institutes and centers. These data cover large area of the Egyptian territory, but they have been taken with different references, instruments, accuracies, and methodologies. At the same time, satellite missions concerning the gravity field of the earth started to yield global gravity models free from the mentioned defects of the terrestrial observations. Recently, many global geoid models have been computed based on satellite data. Meanwhile global satellite only models still suffer from the resolution issue. Those global models need to be verified, investigated, and improved using terrestrial (GPS/Leveling) data. Therefore, this thesis considers two main parts. The first one is evaluating the whole released, until 2019, satellite-based data global models against GPS/Leveling data. The second part is improving some of the best resulted models. EGM2008 and two low d/o satellite data models are also improved using the proposed improving method.Concerning the first part; ninety-three satellite data based global models are evaluated against three GPS/Leveling data sets. EGM2008 as the famous common global model is also evaluated. The results showed that the best ten satellites-based data global models have data either from Goce only or Goce plus other missions like Grace and SLR. Three of them are Goce only data, three of them are (Goce + Grace), and four of them are (Goce + Grace + SLR). The best ten models have d/o ranges from 240 to 300. They considered recent models released from 2014 to 2019. The results showed also that the mean value of the undulation differences ranged from 53 to 56 cm for all the ten models while it was 63 cm in the case of EGM2008. The standard deviation values are ranged from 24.2 to 28.1 cm while it is 25.8 cm in the case of EGM2008. The best ten models have close results to each other and most of them are better than EGM2008 in the test areas.Concerning the second part, three of the ten best models are subjected to the improvement process. Two methods are applied, one is the traditional first order polynomial and the second is proposed simple shifting method. Regarding that the GPS/leveling points in the three data sets have a longitudinal extension, the results showed that the proposed simple shift method with the three data sets gave around 8 cm as a mean and 11 cm as a standard deviation using one seventh of all data points corresponding to 15, 30, and 25 km as radii around the center point in the three data sets, respectively. The results of the simple shift method when using one eleventh and one fifteenth of the data points corresponding to 50 and 60 km as radii around the center point are not very far from the results in theprevious notice. The same can be said about the results of the polynomial method but the mean values in the shifting solutions are noticeably smaller than their corresponding values from the polynomial solutions. Finally, the proposed simple shift method improved EGM2008 and other two low resolution models very significantly.