This paper deals with the performance of the cross- correlation function (C
CF) for the case of time series which contain only a limited number of data
. Often, in this case the data span a time interval shorter than the charac
teristic timescale of the underlying signal. Although various papers recent
ly have been dedicated to this subject, most authors limit their approach t
o numerical simulations only, resulting in a lack of generality and no expl
anation of the difficulties associated with the use of the CCF in many prac
tical situations. We show here, on the basis of theoretical considerations,
that the limits of the CCF can be understood from the fact that in the con
struction of this function the low- frequency components of the signals con
tribute more than the high-frequency components. This result explains also
the robustness of the CCF to the broadband noise.