Abstract: This paper studies regularization in airborne gravity downward to the earth surface. First, the relationship between multicollinearity structure in design matrix of ill-posed linear model and its harm to the unknown parameters is analysed. The information if the unknown parameters suffer the harm are extracted by an index called signal-to-noise ratio, which reveals multicollinearity structure to some extent. Second, regularization based on signal-to-noise ratio(SNR) is put forward. The regularization matrix is constructed according to the ratio. The regularization parameter is selected by minimization the mean squared error of roots to the solution. The regularization matrix constructed in this paper need not use additional physics or priori information, so the algorithm provided in this paper gives a new way of regularization when this kind of information is short of. At last conclutions are drawn from numerical experimerts: compared with ordinary regularization(OR), SNR is better in noise filtering and information fidelity.