Singular points (SPs) in fingerprint images have important applications in fingerprint recognition such as registration and classification. Reliable detection of SPs could sufficiently improve processing speed and performance of automatic fingerprint identification system. In this paper, an improved SPs detection method is proposed. We first present a new algorithm of fingerprint image segmentation using linear support vector machine. Then, we estimate the complex orientation field by multiple scales smoothing in the highest resolution and perform several heuristic rules to enhance the certainty map after complex filtering. Experiments conducted on FVC2002 databases demonstrate that the proposed method can detect SPs with high robustness and accuracy.