The problem of estimating a pdf from measurements has been widely studied by many researchers. However, most of the work was focused on estimating a probability density function of continuous random variables, especially in the absence of noise. In this paper, we consider a model for representing discrete probability density functions based on multirate dsp models. Using this model, we propose an efficient and stable scheme for pdf estimation when the measurements are corrupted by independent additive noise. This approach makes use of well-known results from multirate dsp theory, especially that of biorthogonal partners. Simulation results are given, which clearly show the advantage of the proposed method.