To achieve accurate interferometric synthetic aperture radar (SAR) phase estimation, it is essential to select appropriate high-coherence interferometric pairs from massive SAR single-look complex (SLC) image data. The selection should include as many high-coherence interferometric pairs as possible while avoiding low-coherence pairs. By combining coherence and spectral clustering, a novel selection method for SAR interferometric pairs is proposed in this paper. The proposed method can be adopted to classify SAR SLC images into different clusters, where the total coherence of interferometric pairs in the same cluster is maximized while that among the different clusters is minimized. This is implemented by averaging the coherence matrices of representative pixels to construct an adjacency matrix and performing eigenvalue decomposition for estimating the number of clusters. The effectiveness of the proposed method is demonstrated using 33 TerraSAR-X and 38 dual-polarization Sentinel-1A data samples, yielding improved topography and deformation monitoring results.