Abstract: The atmospheric propaganda delays extracted from multiple reference station GPS networks can server for the fields of weather forecasting, precise positioning and so on, the related extracting method however consists of drawbacks that affect the reliability of the obtainable results, such as inadequate adoption of the known information, severe atmospheric modeling errors. Starting from the modified strategy for GPS network processing, a novel extracting method is derived, whose main points include: (1) directly processing of the un-differenced GPS observables, and elimination of the rank-deficiencies in the observation equations based upon re-parameterization; (2) exploiting the conditions of a-priori known coordinates of the reference stations and the integer-valued nature of the double-differenced ambiguities; (3) introducing the slant ionospheric delays as one of the unknowns, whose temporal behaviors are characterized as constant-velocity process. Experimental results demonstrate that, the short-term predicting precision of the atmospheric delays derived from the new method is at the level better than 1 cm. As the world-widely construction of the Continuous Operating Reference System (CORS), the perspectives of applying the proposed new method can be more and more expanding.