A stochastic model for simulating mountain beef cattle systems was described and validated using real data. The model was able to properly simulate complex batches of animals with different initial characteristics and so obtain results about the mean and variance of animal productive and reproductive outputs. Model was used to simulate four feeding strategies during winter period in the conditions of autumn calving in the Spanish Pyrenees. The simulated strategies were CONTROL: 100% of recommendations for the mean animal; LOW1: as 24% restriction from the amount of energy provided to CONTROL, applied only during lactation; LOW2: the same of LOW1, but was applied only the last two thirds of lactation; LOW3: 16% restriction from CONTROL applied at late gestation and first third of lactation. Simulated results showed that restrictions around 25% of the total winter energy requirements of beef cows in housing conditions are possible and sustainable. The best option taking into account reproductive performance and outputs variability seems to be a continuous restriction during all lactation (strategy LOW1). On the other hand, an extensification policy implying extended grazing till mid or late December with housing afterwards (strategy LOW3), is also possible and will have no negative consequences on calf growth (only viability of young calves being outdoors in autumn should be evaluated), but can affect the PPA of cows and so the herd reproductive performance could be compromised in the long term.