Pansharpening refers to the super resolution of a low-resolution multispectral (LR-MS) image in virtue of an aligned panchromatic (PAN) image. Such an inverse problem mainly requires a proper use of the spatial information from the auxiliary PAN image. In this paper, we suggest a nonconvex regularization model for pansharpening via framelet sparse reconstruction, called NC-FSRM, which investigates the coefficient similarity among the underlying high-resolution MS (HR-MS) and PAN images on the framelet domain, then characterizes the strong statistical sparsity of their error using $ \ell_0 $ norm. Compared with previous methods, NC-FSRM can more precisely and concisely establish the relation between the underlying HR-MS and PAN images. In particular, the piece-wise smoothness prior of the former can simultaneously be captured without adding additional regularizers. For solving the suggested nonconvex model, we further develop an efficient proximal alternating minimization (PAM) based algorithm, which is theoretically proven to converge to the coordinatewise minimizers under some mild assumptions. Numerical experiments conducted on different datasets demonstrate the superiority of the suggested NC-FSRM compared with other state-of-the-art pansharpening methods. The source code is publicly available at https://github.com/zhongchengwu/code_ncFSRM.