Renewable sources, such as wind and solar power, play an increasingly important role in the energy system. If on one hand in many countries there are economic incentives to encourage these types of energies, on the other hand it is required that the power production prediction carried out by system operators is performed as correctly as possible, even introducing forms of economic penalties related to the energy imbalance. It is more and more necessary to be able to produce power production forecasts at least for one-two days ahead with a high degree of accuracy, and provide a measure of the reliability of the same. For solar power production, the irradiance forecasts for two-three days ahead can be performed only by using a numerical model forecast (NWP), while there are several methods to calculate the power production, for example by using physical models, or neural networks (NN). In this study a method based on Analog Ensemble was considered. This algorithm selects some past numerical predictions more similar – according to a given metric – to that of the day considered, and associates the measurements identified through the time of the selected forecasts. In this way it is possible to produce, besides a mean value, a probability distribution function with which estimate the degree of reliability of the forecast. A PV forecast system has been implemented and tested on two Italian sites, applying the Analog Ensemble on the NWP (RAMS) outputs. The performance of the method was compared with a NN method.