To use a liquid-filled portal imaging device (EPID) for transmission dosimetry, it is necessary to understand its dosimetric properties. Therefore, the relation between the pixel values (i.e., ionization currents) of an electronic portal imaging device and the dose rate measured with an ionization chamber in a mini-phantom was investigated. First, a model was introduced to describe the ionization current of the matrix of liquid-filled ionization chambers for pulsed radiation. With this model the relation between ionization current and dose rate is explained qualitatively. Next, buildup measurements were performed at different photon beam energies to assess the amount of buildup material required to obtain electronic equilibrium in the detector. This additional buildup material significantly decreased the image quality, which can hamper patient setup verification, at only the 25 MV beam. Pixel values were then compared with measurements made with a Farmer-type ionization chamber in a mini-phantom at various dose rates. In addition, the influence of a number of accelerator and EPID settings (photon beam energy, pulse rate frequency, gantry rotation angle, and image acquisition modes) on the pixel value was investigated. Subsequently, the dose response relationships of three commercially obtained EPIDs of the same type were compared. For all types of measurements the relation between ionization current and dose rate is described within 1% (1 SD) by an equation with two terms: one term proportional to the square root of the dose rate and another term linear to the dose rate. For images obtained under a typical clinical situation (applying the "normal" acquisition mode at an 8 MV beam with a pulse rate frequency of 400 Hz at a transmission dose rate of 100 cGy/min) the contribution of the square root and linear term to the EPID signal is 94% and 6%, respectively. The weight factors of both terms depend on the photon beam energy, pulse rate frequency, and image acquisition mode. It is concluded that the EPID is useful for dosimetry purposes with 1% (1 SD) accuracy, but that the dose response relationship has to be determined for each EPID and accelerator setting.