Many researchers install Campbell-Stokes sunshine recorders to evaluate the global irradiation from heliophany measurements using the well-known Angstrom linear regression. Because of the statistical nature of this relation, the actual solar irradiation values are determined with some given standard deviation, the fluctuation of which a poor empirical solar radiation database can contribute to. In this paper a procedure (＇＇cross-validation＇＇) has been devised, based on 8 years of daily sunshine and daily global solar radiation data, in order to investigate the dependence of the general standard deviation on the size of the database. Therefore, one can infer that a database of at least 14 years is necessary and may be sufficient not only to obtain, by Angstrom regression, a trustworthy and statistically stable estimation of the daily global solar irradiation but, further, to set up a ＇＇theoretical＇＇ and statistically stable distribution of any climatic quantity. Beyond this limit, the variability of the climatological quantities originates from only their intrinsic statistical nature, independently of the size of the database used.