Transformation of ellipsoidal heights derived from the Global Positioning System (GPS) to orthometric heights using geoid models is investigated in the north and west parts of Turkey. Although the transformation depends on a simple relation between ellipsoidal h, orthometric H and geoid N heights, the accuracy of the resulting orthometric heights after transformation is crucial in geodetic and surveying applications. Various factors which affect this accuracy, such as measurement errors, datum inconsistencies and theoretical assumptions, are investigated in this study, while testing different methods in three test networks (Sakarya in the Northwest, Cankiri in the North and Izmir in the West). The study consists of three steps. In the first step the regional Turkey geoids TG99A, TG03 and the European gravimetric geoid EGG97 are tested comparing geoid heights derived from models and GPS/levelling at co-located benchmarks. In the second step, regional geoid models are combined with GPS/levelling using Least Squares Adjustment of height differences and corrector surface models. In this step, additionally, Variance Component Estimation (VCE) using Minimum Norm Quadratic Unbiased Estimation (MINQUE) approach is performed, in order to combine the height sets. In the last step, GPS/levelling surface type local geoids are determined and their performances are tested in transformation of GPS-heights. Finally, the resulting accuracies are compared and practical aspects of those approaches in deriving orthometric heights from GPS measurements in geodetic and surveying applications are discussed.