In our clinic a QA program for IMRT verification, fully based on dosimetric measurements with electronic portal imaging devices (EPID), has been running for over 3 years. The program includes a pre-treatment dosimetric check of all IMRT fields. During a complete treatment simulation at the linac, a portal dose image (PDI) is acquired with the EPID for each patient field and compared with a predicted PDI. In this paper, the results of this pre-treatment procedure are analysed, and intercepted errors are reported. An automated image analysis procedure is proposed to limit the number of fields that need human intervention in PDI comparison.Most of our analyses are performed using the gamma index with 3% local dose difference and 3mm distance to agreement as reference values. Scalar parameters are derived from the gamma values to summarize the agreement between measured and predicted 2D PDIs. Areas with all pixels having gamma values larger than one are evaluated, making decisions based on clinically relevant criteria more straightforward.In 270 patients, the pre-treatment checks revealed four clinically relevant errors. Calculation of statistics for a group of 75 patients showed that the patient-averaged mean gamma value inside the field was 0.43 +/- 0.13 (1SD) and only 6.1 +/- 6.8% of pixels had a gamma value larger than one. With the proposed automated image analysis scheme, visual inspection of images can be avoided in 2/3 of the cases.EPIDs may be used for high accuracy and high resolution routine verification of IMRT fields to intercept clinically relevant dosimetric errors prior to the start of treatment. For the majority of fields, PDI comparison can fully rely on an automated procedure, avoiding excessive workload.