Previous studies have used low-resolution models to investigate the potential predictability of river discharges. The present study conducts a preliminary investigation of this predictability based on a dynamical seasonal prediction method using an atmospheric general circulation model (AGCM) and studies its dependency on the AGCM’s horizontal resolution. I performed 60-km and 180-km mesh AGCM simulations forced with the observed sea-surface temperatures and sea-ice concentration and simulations performed by converting the 60-km mesh simulation results to 180-km mesh ones and vice-versa. The global geographical distributions of the potential predictability resemble those of previous studies; in addition, the river basin correction effect is reconfirmed and found in basins exceeding 104km2 in area. Comparisons of the potential predictability among the multiple simulations demonstrate that upscaling loses the detailed features of the potential predictability, while downscaling retains the details. Dominant sources of the difference in the potential predictability of river discharge between the 60 km and 180 km mesh AGCMsimulations are their atmospheric potential predictabilities.