The thesis describes the use of a liquid-filled electronic portal imaging device, EPID, for in vivo dosimetry, i.e., for the verification of transmission, exit, and midplane dose. First, the application of the EPID to measure the transmission dose, i.e., the dose behind the patient, was investigated under various clinical conditions. For this purpose corrections for the nonlinear response of the EPID with dose and the effects of photon scatter in the EPID were performed. Next, a convolution model was developed to convert transmission dose images to exit and midplane dose distributions in the patient. Patient inhomogeneities are taken into account by using the radiological path length through the patient, which is derived from the measured transmission dose, i.e., without using patient data (CT). Finally, the accuracy of EPID exit and midplane dose measurements was determined for several homogeneous and inhomogeneous phantoms in open and wedged beams and for a few patients treated for larynx, breast, lung and prostate cancer. It was found that for most phantoms exit and midplane doses agree within 2% (1 s.d.) with ionization chamber measurements. EPID midplane dose measurements during patient treatment generally agreed within 2.5% with 3D treatment planning calculations. Occasionally larger differences were observed due to differences between the actual patient anatomy during treatment and the patient CT data, used for treatment planning. It is concluded that transmission dosimetry with a liquid-filled EPID can be used to determine the 2D exit and midplane dose for several treatment sites in clinical practice and to quantify the influence of anatomical changes on the dose delivery.