In-vivo dosimetry (IVD) in external beam radiotherapy is used to detect major clinically relevant differences between planned and delivered dose. Moreover, a detailed analysis of its results, when routinely reported and discussed by the radiotherapy staff, can limit the likelihood of error transmission to many treatments. A first experience of routine EPID-based IVD in a reference point has been performed in our department for 3D-CRT treatments over a three-year period. More than 14,000 images were acquired and 1287 treatment plans were verified. The IVD checks were obtained three times in the first week and then weekly. Tolerance levels of ± 5% for pelvic-abdomen, head-neck and breast irradiations and ± 6% for lung treatments were adopted for the in-vivo measured dose per fraction. A statistical analysis of the IVD results was performed grouping the data by: anatomical regions, treatment units, open and wedged fields and gantry angles. About 10% of the checked doses per fraction showed dosimetric discrepancies out of the tolerance levels. The causes of the discrepancies were 70% delivery or planning errors, 20% morphological changes and 10% procedural limitations. 41 cases (3.2%) have required special investigations because their in-vivo doses per fraction, averaged over the first three sessions, were out of the tolerance levels and in 19 cases (1.5%) the deviations gave rise to an intervention. Statistically significant differences of average variations between planned and delivered doses were observed for: (i) 30° wedged 10 MV fields with respect to those of other wedged or open 10 MV fields delivered by two linacs, due to the incorrect TPS implementation of that wedge transmission factor; (ii) anterior-posterior and posterior-anterior beams with respect to the other gantry orientations for one linac, due to the beam attenuation introduced by the treatment couch; (iii) lateral fields with respect to medial fields of breast irradiations for all linacs, due to small systematic set-up variations. The analysis of our data shows a substantial homogeneity of the IVD results for all the considered body regions and treatment units. However, the observed discrepancies have supplied indications for taking further steps in the optimization process and in some cases to adopt an adaptive approach.