In a view to make easier the use of meterological data as the input of energy systems simulations, this study proposes a methodology to reduce a long time series, of several variables into some typical meteorological days, characterizing classes with their frequency distributions and transition probabilities. To do it, we have defined parameters elaborated from hourly clearness index profiles computed from hourly horizontal and inclined irradiation recorded at two nearest Mediterranean coastal sites in Corsica Island. The classification used the aggregation Ward＇s method, and the results are confirmed by discriminant analyses. Two studies based on different time steps (monthly and annual) have been investigated. The classes are more compact, and the variances of the meteorological and mathematical variables in a given class remain quite low by comparison with the mean. We show in the first part (monthly time step) that the recorded days are clustered in 3, 4 or 5 groups and in the second part (annual time-step on about 4 years of irradiation data), 3 typical meteorological days classes allows to characterize the irradiation profiles of both locations. (C) 2000 Elsevier Science Ltd. All rights reserved.