Abstract: In this paper, for about one monthly water level data taken respectively from Yah Zhou and Junan deep wells, the water level is not used as input channel into wenner filter so that the disturbing facters suchas air pressure and solid tide can be removed and the weak anomalous information on water level rising and drop before an earthquake can be extracted.The calculation results show that this method can rais signal-to-noise ratio. Taking the signal-to-noise ratio which is two times larger than standard deviation as an anomalous criterion, the resolution for an anomalous point of 2 cm or more is up to 61% and 96%, for an anomalous point of 3 cm or more is to 86% and100%.The random property of the residual curve is also obviously decreased. Compared with the method in which water level is used as input channel into Wenner filter, this method raised the anomlous resolution from original 4-9% to 61-100%. Compared with gradual regression analysis method, it remarkably improved the radom property of residiual curves. It is especially suitabte to Junan well in which solid tide difference is large and solid tide coefficient difference of diurnal wave and semi-diurnal wave is large; the anomalous resolution is also relatively raised. This paper also, in terms of filter operator, discusses the reason why the solid tide coefficient difference letwen diurnal wave and semidiurnl wave in Junan well is comparably large.