Conservation actions that have maintained populations in the past may not necessarily do so in the future. Population viability analysis provides one tool for exploring the impact of management actions on large temporal scales. However, there are relatively few long-term data sets that provide the demographic and environmental data demanded by such models. Using a 37-yr data set, we used RAMAS Metapop to model the persistence of natterjack toads Epidalea (Bufo) calamita on a heathland in southern Britain. A retrospective analysis showed that the best fit between the predicted population trajectories and the real population was when the management carried out was modelled as an increase in K of 150 toads yr−1. However, even if ongoing management continues to improve K by a further 40–60 toads yr−1 over the next 50 yr, the population still has an extinction risk of at least 60% if other factors remain unchanged. Sensitivity analyses and simulated management scenarios indicated that the population was most sensitive to changes in the survival of juvenile (i.e. 1–2 yr old) toads. In addition, if the frequency and severity of pond desiccation increases, the risk of extinction was predicted to increase as a result of reduced recruitment. Low levels of extinction risk occurred irrespective of K when juvenile survival was enhanced in combination with low frequency and severity of pond desiccation. The models suggest that populations that are responding to management against a background of natural fluctuations may remain vulnerable to extinction for several decades. These extinction risks may increase if habitat management fails to offset reductions in recruitment and juvenile survival caused by environmental change.