Abstract: Developing the blind signal processing method in wireless sensor networks (WSNs) often needs to consider several constraints including limited communication bandwidth and energy of sensors. The processing of observed information will import a variety of quantization noise, which is always difficult to be modelled accurately by simple probabilistic models. To study the extraction issue of signal with unknown statistics in WSNs, a signal extracted method based on a cost reference particle filter (CRPF) is proposed in this paper. The method attains the accuracy of prediction particles by cubature-points transformation, and completes particles updating and propagation through cost-risk functions. Simulation results show that the proposed method has comparable performance with the other algorithms for noise of unknown statistics.