Two complementary approaches to the noise suppression problem in on-line portal imaging have been analysed. Temporal filtering by image summation can substantially reduce the amount of noise in an image. In many cases, however, movements of the patient or the radiation source limit the time period over which the averaging can be done. Any remaining noise has to be dealt with by applying spatial filtering. The adaptive Lee filter is particularly suitable for portal imaging applications. It preserves a crisp definition of edges while removing noise in flat regions of the image. It can be used to obtain images of satisfactory quality with short radiation exposure of the patient. We have proposed a modification to the basic Lee technique which permits the calculation of the noise variance locally by utilising the information contained in intermediate images acquired during frame averaging. Unlike the original Lee formulation, no a priori knowledge of the noise variance is required, and in contrast to Mastin's approach (Mastin 1985), the variance may vary with position in the image. The tests of performance of the modified Lee filter, carried out using on-line images, have shown its superiority in comparison with the original Lee technique as well as with conventional averaging and median filters.