Abstract In order to elaborate a generator of “typical days” sequences necessary for sizing photovoltaic (PV) systems, this paper develops a methodology to class daily global solar irradiation based on a first order Markov chain model. The method has been applied for two stations in Corsica (France), and we have limited the number of discriminant parameters computed from the hourly clearness index kt ( h ). The classification is based on Ward’s method, checked by discriminant analyses. Previous works have shown that the different days could be clustered in three groups with distinct mean values and lower standard deviations. After checking the dependence and the stationary of the Markov’s chain, the presented model allows one to compute the simulated marginal probabilities p i with a good correlation with the experimental ones (MBE ∈ [−0.154;0.151]; mean bias error, MBE). A good correlation has been observed between the simulated and experimental transition probability matrices for both meteorological sites (Vignola MBE=0, RMSE ∈ [0.035;0.056], RMBE(%) ∈ [0.17;3.59]; Campo del’Oro MBE=0, RMSE ∈ [0.016;0.105], RMBE(%) ∈ [−2.15;4.26]). At least, the sizing of PV systems has shown that the kW h costs, computed from real and simulated systems, present weak relative errors belonging to the range −13.3% to +4.2%, checking the global efficiency of the methodology.