The optical losses associated with the attenuation of the reflected direct irradiance by the heliostats along the optical path to the receiver may be significant in large solar tower plants. This phenomenon, known asatmospheric attenuation loss, may have a stronger impact at those tower plants where high aerosol loads are expected. Performance models like the System Advisor Model (SAM) and the ray-tracing models (DELSOL, MIRVAL) usually estimate the atmospheric attenuation loss by a polynomial expression which is function of the slant range (the optical path between the heliostat and the receiver). Most of the polynomial models proposed to determine this optical loss use two established extreme attenuating conditions corresponding to a clear or hazy atmosphere. This paper presents a sensitivity study of the impact of time-dependent variability of the atmospheric attenuation in the yield performance of two reference large solar tower plants (one similar to Ivanpah 1 and the other one to Crescent Dunes as examples of direct steam and molten salt tower plants, respectively). Five sites have been selected from the AERONET ground station network to obtain the aerosol loading at different time-scales: annual, monthly and daily. Multiple SAM runs have been performed to simulate the annual yield of each plant and site creating different inputs to the code corresponding to each time-scale condition. The results show a significant impact of the time-scale for modeling the atmospheric attenuation on the annual yield and daily energy output of the plant. Although the annual and monthly means produce some compensation of the impact, the differences in several particular days can be significant. Up to 20% difference in the daily energy output is found when the extinction is modeled asa steady-state polynomial representing the annual mean compared to the case of daily time-dependent variability of the attenuation. The sensitivity results presented here show that for more realistic yield performance calculations in solar tower plants, particularly at desert and arid climates, the modeling of the atmospheric attenuation should be performed in a time-dependent way according to the climatological variability conditions characteristic of the site.