Insufficient filtration and over-smoothing are misleading processes in the quantification of time-activity curves. The optimum filtration requires a good knowledge of the frequency spectrum and relative amplitudes of the data and superimposed noise. Due to variations in biomedical data, it is very difficult to adjust the filter for individual cases. To overcome this problem a new method of noise reduction is proposed. In this method the time-activity curves are transformed into a low frequency (linear) curve that can be filtered heavily without significant distortion of the real data. The theory of the proposed filter and the results of its comparison with three-point filter, five-point filter and data bounding methods are presented. The comparison was performed using deconvolution analyses of simulated renograms. The results show that the proposed filter causes minimum distortion of the renogram and impulse retention function in terms of the root mean square error and the peak of the renogram. Moreover, the filter is much less sensitive to over-smoothing (number of filter iterations), the signal-to-noise ratio and the mean transit time of the renogram compared with other filters.