Clouds are a major factor in the solar radiation transmittance throughout the atmosphere and have a large impact in the availability of the solar energy resource. A variety of radiative transfer models estimate the surface solar irradiation from an interpolation between clear sky and of overcast conditions, using the effective cloud cover as modulation factor. The effective cloud cover parameter can be calculated from satellite data, considering the visible radiances attributed to clear sky and overcast conditions and the actual radiance detected by the satellite at the time of the estimation, where the visible radiance attributed to overcast conditions is frequently estimated as the maximum visible radiance detected by the satellite over a previous period, or a predetermined fraction of that value. However, very high visible radiances can be frequently observed in tropical regions, associated with clouds with high vertical development. For those regions, to estimate the satellite visible radiance associated to overcast conditions as a function of the maximum radiance can lead to overestimation of this parameter and subsequently to errors in the assessment of the available surface irradiance, as well as in errors in the direct/diffuse irradiance partition. This work aimed to improve the assessment of the solar energy resource in tropical regions, by estimating the satellite visible radiance most likely to correspond to overcast conditions on a regional scale using instead surface measurements of Direct Normal Component [DNI]. Statistical methods were applied to four years of [DNI] observations in 22 stations distributed across Brazil, and the GOES-13 satellite visible radiances corresponding to the pixel of the station at the time of the observation. Visible radiances were corrected by the solar zenith angle (limited to values less than 60 degrees). The hypothesis that a threshold visible radiance can separate overcast conditions (here, those with DNI < 50 W) from clear or partially cloudy sky conditions, was tested against the corresponding [DNI] data in the pixel, resulting in a contingency table for each threshold. The Peirce Skill Score [PSS], Threat score [TS], and different combinations of the Precision and Recall rates given by the Fβ score, with β=0.5, 1 and 2, were calculated for each possible threshold. Values of visible radiance between 10 and 1000 counts, with an interval of 10 counts, were considered as possible thresholds. Optimal values (e.g. those that maximize the index in the series) ranged between 190 and 350 counts (depending of the skill score considered), 4 to 6 times lower than the maximum radiance values typically observed in the region. Differences between stations were obtained, but a geographical pattern was not observed. Results were tested in Brasil-SR radiative transfer model and differences in global and direct irradiance estimations are discussed. The method presented here is applicable to the estimation of overcast conditions using data from different satellites and geographical locations, and could be used by other models that use visible satellite data for the estimation of the effective cloud cover. This work is a contribution of the National Institute of Science and Technology for Climate Change INCT-MC, financed by FAPESP 2014/50848-9, CNPq 465501/2014-1, and CAPES/FAPS Nº 16/2014.