This paper presents a methodology to predict in a statistically reliable way the behavior of a grid-connected photovoltaic system. The methodology developed can be implemented either in common programming software or through an off-the-shelf simulation of electrical systems. Initially, the atmospheric parameters that influence the behavior of PV generators (radiation and temperature) are characterized in a probabilistic manner. In parallel, a model compound by various PV generator components is defined: the modules (and their electrical and physical characteristics), their connection to form the generator, and the inverter type. This model was verified for comparing their behavior with output measured on a real installed system of 3.6 kWp. The solar resource characterized and the photovoltaic system model are integrated in a non-deterministic approach using the stochastic Monte Carlo method, developed in the programming language DPL of the electrical-systems simulation software DIGSILENT(R). It is done to estimate the steady-state electrical parameters describing the influence of the grid-connected photovoltaic system. Specifically, we estimated the nominal peak power of the PV generator to minimize network losses, subject to constraints on nodes voltages and conductor currents. (C) 2012 Elsevier Ltd. All rights reserved.