Purpose The aim of this work was to evaluate the SunCHECK PerFRACTION, the software for in vivo monitoring using EPID images. Materials/Methods First, the PerFRACTION ability to detect errors was investigated simulating two situations: (1) variation of LINAC output and (2) variation of the phantom thickness. An ionization chamber was used as reference to measure the introduced dose variations. Both tests used EPID in integrated mode (absolute dose). Second, EPID measurements in integrated mode were carried out during an independent Brazilian governmental audit that provided four phantoms and TLDs. PerFRACTION calculated the absolute dose on EPID plane, and it compared with predicted calculated dose for every delivered plan. The dose deviations reported using PerFRACTION were compared with dose deviations reported by the independent audit. Third, an end‐to‐end test using a heterogeneous phantom was performed. A VMAT plan with EPID in cine mode was delivered. PerFRACTION calculated the mean dose on CBCT using EPID information and log files. The calculated doses at four different points were compared with ionization chambers measurements. Results About the first test, the largest difference found was 1.2%. Considering the audit results, the variations detected by TLD measurements and by PerFRACTION dose calculation on EPID plane were close: 12 points had variations less than 2%, 2 points with variation between 2% and 3%, and 2 points with deviations greater than 3% (max 3.7%). The end‐to‐end tests using a heterogeneous phantom achieved dose deviation less than 1.0% in the water‐equivalent region. In the mimicking lung region, the deviations were higher (max 7.3%), but in accordance with what is expected for complex situations. Conclusion The tests results indicate that PerFRACTION dose calculations in different situations have good agreement with standard measurements. Action levels were suggested for absolute dose on EPID plane as well as 3D dose calculation on CBCT using PerFRACTION.