Nuclear counting statistics at high count rate are assessed on a -ray spectrometer set-up with a Wilkinson analog to digital convertor. The validity of recent theoretical formulas for the standard deviation, before and after pulse-loss compensation, is checked. The experimental counting uncertainty is well reproduced by theory. Without pulse-loss compensation (cf. real-time mode), it is dependent on the size and position of the considered region of interest in the spectrum. With pulse-loss compensation (cf. live-time correction) the relative deviation from Poisson statistics is equal for all regions of interest in the spectrum.