Representative meteorological data from a given location are necessary to assess the long-term performance of photovoltaic (PV) systems. Test reference years (TRYs) or typical meteorological years (TMYs) are widely used as input to PV models. Most of current procedures propose the construction of TRYs by concatenating 12 months belonging to different years of a dataset. This paper evaluates the effects of the temporal downscaling of typical periods that compose different TRYs on the long-term assessment of PV systems. The Festa-Ratto TRY, WYSS, EN ISO 15927-4 TRY, TMY3, TGY and TDY are used. Thus, an adapted version of these six methodologies aimed at the selection of typical days rather than months is proposed. The electricity production obtained by simulation for daily and monthly TRYs is compared with simulations performed for each actual year of the dataset. This analysis is performed for seven locations in the USA considering a 5.6 kWp grid-connected PV system. The results reveal that the timescale reduction improves the behavior of Festa-Ratto TRY, WYSS, TMY3, TDY and TDY when estimating the long-term production of a PV system considering the hourly, daily, monthly and annual timescales, while the modified EN ISO 15927-4 TRY performs worse than its monthly version. (C) 2018 Elsevier Ltd. All rights reserved.