Nous posons le probl`eme de l’estimation de l’irradiance comme un probl`eme d’estimation statistique et en calculons la borne de performance (borne de Cram´er-Rao). Nous montrons qu’elle ne peut ˆetre atteinte, mais qu’une m´ethode r´ecente [5] permet de l’approcher. Nous montrons ensuite qu’en tenant compte des ´echantillons satur´es, les r´esultats d’estimation de l’irradiance sont am´elior´es. Les r´esultats obtenus justifient l’introduction d’un nouvel estimateur qui, contrairement aux m´ethodes de la litt´erature, prend en compte les ´echantillons satur´es. We state the problem of irradiance estimation as a statistical parameter estimation problem and compute the performance bounds of the estimation (Cram´er-Rao lower bound). We show that no efficient estimator exists under the stated hypothesis. However, we find that a recently proposed method [5], performs very close to this bound. Moreover, we show that releasing one of the working hyphotesis (including the saturated samples) the irradiance estimation can be improved. We introduce a new estimation method that, unlike existing methods, takes into account the information provided by the saturated samples