Radiative transfer model calculations of solar fluxes during cloud-free periods often show considerable discrepancies with surface radiation observations. Many efforts have been undertaken to explain the differences between modeled and observed shortwave downward radiation (SDR). In this study, MODTRAN4v3r1^TM (designed later simply as MODTRAN^TM) was used for model simulations and compared with high-quality radiation observations of the Baseline Surface Radiation Network (BSRN) site at Payerne, Switzerland. Results are presented for cloud-free shortwave downward radiation calculations. The median differences of modeled minus observed global SDR are small (<1%) and within the instrumental error. The differences of modeled and observed direct and diffuse SDR show larger discrepancies of −1.8% and 5.2%, respectively. The diffuse SDR is generally overestimated by the model, and more important, the model to observation linear regression slope and zero intercept differ significantly from their ideal values of 1 and 0. Possible reasons for the discrepancies are presented and discussed, and some modifications are investigated for decreasing such differences between modeled and observed diffuse SDR. However, we could not resolve all the discrepancies. The best agreement is obtained when comparing model simulations whose 550-nm aerosol optical depth input is inferred from observations using nine spectral channels and using BSRN observations performed with a new and more precise shading disk and Sun-tracking system. In this case, the median bias between model simulations and observed diffuse SDR is −0.4 W m⁻² (<1%).