As a classic eye movement method, electrooculogram (EOG) has been extensively used in many applications. There are many different types of eye movements and artifact in the EOG signal. Noise attenuation and signal separation have received special attention in the EOG research. In this paper, we introduce a novel Linear-nonlinear combinational filter based on weighted FIR-median-hybrid (WFMH) with the characteristic of the EOG signal. The result of the simulation shows that this filter has the property of removing random noise more efficiently when preserving sharp edges. Finally, it is shown that the new filter is effective in separating saccadic and eye blink in the EOG signal.