The paper reports the results of a photogrammetric survey made using an Unmanned Aerial Vehicle (UAV) in the archaeological site of the Roman Amphitheatre in Avella (Avellino, Italy). The aim of the study is to verify which modality of image acquisition (if only nadiral images or nadiral plus Oblique images), together with the method of Global Positioning Satellite System (GNSS) survey of the Ground Control Points (GCP) is able to produce the better 3D model, in terms of accuracy, in order to extract traditional graphic drawings (plan, elevation and section), suited to the required representation scales (1 : 100 and 1 : 50). The accuracy in georeferencing was evaluated analysing the residues on the GCPs; subsequently, a more detailed analysis of the accuracy of the final 3D model was performed analysing the residuals on the image coordinates, also called re-projection error. The method developed is based on the statistical analysis of the different models, built changing the GCPs survey method and the photogrammetric shots acquired. The results of our analysis show that the photogrammetric survey is more ‘stable’ using only nadiral images and that the nRTK technique allows results comparable to those obtained with static measurements, both in precision and in reliability. Moreover, if the GCPs are measured in nRTK mode, taking into consideration the graphical error, the maximum representation scale is 1 : 100, whereas the use of static technique makes it possible to describe major details, at a scale of 1 : 50.