To extract singular points from low quality fingerprint image, a two-phase algorithm for singular points extracting is presented in this paper. In the first phase, an improved Poincare index is used to extract candidate singularities. In the second phase, Gaussian-Hermite moment attribution of ridge for each candidate is calculated to determine its authenticity. Because of effectively assembling information of ridge orientation and coherence in singularity’s neighborhood, this method can extract singularities in a relatively accurate and reliable way. Experiments have shown its effectiveness and robustness.