Solar radiation is characterized by short fluctuations introduced by passing clouds. An analysis of these fluctuations with regard to solar energy applications should focus on the instantaneous clearness index. Its probability distribution for a given mean clearness index is, as a first approximation, independent from the season and partly also from the site. This is verified for four annual datasets from three different sites. An analysis of fluctuations in solar radiation must focus on their amplitude, persistence, and frequency of occurrence rather than their location in time. The Fourier analysis cannot satisfactorily provide this information since time series of the instantaneous clearness index exhibit no periodicity. Instead, a localized spectral analysis based on wavelet bases rather than on periodic-ones has been applied. This analysis allows the decomposition of the fluctuating clearness index signal into a set of orthonormal subsignals. Each of them represents one specific scale of persistence of the fluctuation. The annual mean square values of all subsignals have been analysed, permitting the allocation of the signal's power content to the different scales of persistence of a fluctuation. These annual mean values agree well for the different datasets, indicating the existence of statistically significant mean square values of the fluctuations as a function of their persistence. The analysis offers a valuable tool for the estimation of power flow fluctuations introduced by direct solar energy systems. With further elaboration it may be applied by power system operators for network planning in distribution grids with a high density of embedded generation. (c) 2006 Elsevier Ltd. All rights reserved.