Inverter sizing strategies for grid-connected photovoltaic (PV) systems often do not take into account site-dependent peculiarities of ambient temperature, inverter operating temperature and solar irradiation distribution characteristics. The operating temperature affects PV modules and inverters in different ways and PV systems will hardly ever have a DC output equal to or above their STC-rated nominal power. Inverters are usually sized with a nominal AC output power some 30% (sometimes even more) below the PV array nominal power. In this paper, we show that this practice might lead to considerable energy losses, especially in the case of PV technologies with high temperature coefficients of power operating at sites with cold climates and of PV technologies with low temperature coefficients of power operating at sites with warm climates and an energy distribution of sunlight shifted to higher irradiation levels. In energy markets where PV kW It's are paid premium tariffs, like in Germany, energy yield optimization might result in a favorable payback of the extra capital invested in a larger inverter. This paper discusses how the time resolution of solar radiation data influences the correct sizing of PV plants. We demonstrate that using instant (10 s) irradiation values instead of average hourly irradiation values leads to considerable differences in optimum inverter sizing. When calculating inverter yearly efficiency values using both, hourly averages and 1-min averages, we can show that with increased time resolution of solar irradiation data there are higher calculated losses due to inverter undersizing. This reveals that hourly averages hide important irradiation peaks that need to be considered. We performed these calculations for data sets from pyranometer readings from Freiburg (48 degrees N, Germany) and Florianopolis (27 degrees S, Brazil) to further show the peculiarities of the site-dependent distribution of irradiation levels and its effects on inverter sizing. (c) 2005 Elsevier Ltd. All rights reserved.