Most existing edge-detection algorithms are based on the derivatives of potential-field data, and thus, enhance high wavenumber information and are sensitive to noise. The normalized anisotropy variance method (NAV-Edge) was proposed for detecting edges of potential-field anomaly sources based on the idea of normalized standard deviation (NSTD). The main improvement over the balanced, windowed normalized variance method (i.e., NSTD) used for similar purposes was the application of an anisotropic Gaussian function designed to detect directional edges and reduce sensitivity to noise. NAV-Edge did not directly use higher-order derivatives and was less sensitive to noise than the traditional methods that use derivatives in their calculation. The utility of NAV-Edge was demonstrated using synthetic potential-field data and real magnetic data. Compared with several existing methods (i.e., the curvature of horizontal gradient amplitude, tilt angle and its total-horizontal derivative, theta map, and NSTD), NAV-Edge produced superior results by locating edges closer to the true edges, resulting in better interpretive images.