Signal digitisation may produce significant effects in balloon borne or space CMB experiments, when the limited bandwidth for downlink of data requires for loss-less data compression. In fact, the data compressibilty depends on the quantisation step q applied on board by the instrument acquisition chain. In this paper we present a study of the impact of the quantization error in CMB experiments using, as a working case, simulated data from the Planck/LFI 30 and 100 GHz channels. At TOD level, the effect of the quantization can be approximated as a source of nearly normally distributed noise, with RMS ≃ q/ √ 12Ns, with deviations from normality becoming relevant for a relatively small number of repeated measures Ns< ∼20. At map level, the data quantization alters the noise distribution and the expectation of some higher order moments. We find a constant ratio, ≃ 1/( √ 12σ/q), between the RMS of the quantization noise and RMS of the instrumental noise, σ over the map (≃ 0.14 for σ/q ≃ 2), while, for σ/q ∼ 2, the bias on the expectation for higher order moments is comparable to their sampling variances. Finally, we find that the quantization introduces a power excess, C l , that, although related to the instrument and mission parameters, is weakly dependent on the multipole l at middle and large l and can be quite accurately subtracted. For σ/q ≃ 2, the residual uncertainty, ∆C l , implied by this subtraction is of only ≃1–2% of the RMS uncertainty, ∆C l , on C l reconstruction due to the noise power, C l . Only for l ∼30 the quantization removal is less accurate; in fact, the 1/f noise features, although efficiently removed, increase C l , ∆C noise l , C ex l and then ∆C ex l ; anyway, at low multipoles C sky l ≫ ∆C l > ∆C l . This work is based on Planck LFI activities.