Context. The observed presence of pure CO 2 ice in protostellar envelopes, revealed by a double peaked 15 μm band, is often attributed to thermally induced ice segregation. The temperature required for segregation is however unknown because of lack of quantitative experimental data and this has prevented the use of ice segregation as a temperature probe. In addition, quantitative segregation studies are needed to characterize diffusion in ices, which underpins all ice dynamics and ice chemistry. Aims. This study aims to quantify the segregation mechanism and barriers in different H 2 O:CO 2 and H 2 O:CO ice mixtures. Methods. The investigated ice mixtures cover a range of astrophysically relevant ice thicknesses and mixture ratios. The ices are deposited at 16-50 K under (ultra-)high vacuum conditions. Segregation is then monitored, at 40-70 K in the CO 2 mixtures and at 23-27 K in the CO mixtures, through infrared spectroscopy. The CO 2 and CO band shapes are distinctly different in pure and mixed ices and can thus be used to measure the fraction of segregated ice as a function of time. The segregation barrier is determined using rate equations and the segregation mechanism is investigated through Monte Carlo simulations. Results. Thin (8-37 ML) H 2 O ice mixtures, containing either CO 2 or CO, segregate sequentially through surface processes, followed by an order of magnitude slower bulk diffusion. Thicker ices (>100 ML) segregate through a bulk process, which is faster than even surface segregation in thin ices. The thick ices must therefore be either more porous or segregate through a different mechanism, e.g. a phase transition, compared to the thin ices. The segregation dynamics of thin ices are reproduced qualitatively in Monte Carlo simulations of surface hopping and pair swapping. The experimentally determined surface-segregation rates follow the Ahrrenius law with a barrier of 1080 ± 190 K for H 2 O:CO 2 ice mixtures and 300 ± 100 K for H 2 O:CO mixtures. Though the barrier is constant with ice mixing ratio, the segregation rate increases with CO 2 concentration. Conclusions. Dynamical ice processes can be quantified through a combination of experiments and different model techniques and they are not scale independent as previously assumed. The derived segregation barrier for thin H 2 O:CO 2 ice mixtures is used to estimate the surface segregation temperature during low-mass star formation to be 30 ± 5 K. Both surface and bulk segregation is proposed to be a general feature of ice mixtures when the average bond strengths of the mixture constituents in pure ice exceeds the average bond strength in the ice mixture.