We developed the landscape age‐class demographics simulator ( LADS) to model historical variability in the amount of old‐growth and late‐successional forest in the Oregon Coast Range over the past 3,000 years. The model simulated temporal and spatial patterns of forest fires along with the resulting fluctuations in the distribution of forest age classes across the landscape. Parameters describing historical fire regimes were derived from data from a number of existing dendroecological and paleoecological studies. Our results indicated that the historical age‐class distribution was highly variable and that variability increased with decreasing landscape size. Simulated old‐growth percentages were generally between 25% and 75% at the province scale (2,250,000 ha) and never fell below 5%. In comparison, old‐growth percentages varied from 0 to 100% at the late‐successional reserve scale (40,000 ha). Province‐scale estimates of current old‐growth (5%) and late‐successional forest (11%) in the Oregon Coast Range were lower than expected under the simulated historical fire regime, even when potential errors in our parameter estimates were considered. These uncertainties do, however, limit our ability to precisely define ranges of historical variability. Our results suggest that in areas where historical disturbance regimes were characterized by large, infrequent fires, management of forest age classes based on a range of historical variability may be feasible only at relatively large spatial scales. Comprehensive landscape management strategies will need to consider other factors besides the percentage of old forests on the landscape, including the spatial pattern of stands and the rates and pathways of landscape change.