The use of rational functions has become a standard for very high-resolution satellite imagery (VHRSI). On the other hand, the overall geolocalization accuracy via direct georeferencing from on board navigation components is much worse than image ground sampling distance (predicted < 3.5 m CE90 for WorldView-3, whereas GSD = 0.31 m for panchromatic images at nadir). <br><br> This paper presents the georeferencing accuracy results obtained from a single WorldView-3 image processed with a bias compensated RPC camera model. Orientation results for an image collected over Milan are illustrated and discussed for both direct and indirect georeferencing strategies as well as different bias correction parameters estimated from a set of ground control points. Results highlight that the use of a correction based on two shift parameters is optimal for the considered dataset.