This study reports a novel gradient nonlinearity (GNL) calibration approach using DIADEM (Distortion-free Imaging Approach with a Double Encoding Method) diffusion imaging. Unlike standard diffusion-weighted echo-planar-imaging (DW-EPI), DIADEM is free from DW-EPI distortions. This allows GNL calibration with a uniform phantom, since confounding effects between DW-EPI and GNL-induced distortions in the calibration are separated. Direct bias correction could be applied to the corresponding in-vivo data from the DIADEM scans, which results in reliable quantitative diffusion imaging. The feasibility was successfully demonstrated in phantom and in-vivo on a compact 3T system.