We developed dynamic bioenergetics models to investigate how behavioural responses to anthropogenic disturbance events might affect the population dynamics of three marine mammal species (harbour porpoise, grey seal and harbour seal) with contrasting life‐history traits (capital versus income breeders) and movement behaviour (resident versus nomadic). We used these models to analyse how individual vital rates were affected by differences in the probability of disturbance and the duration of any behavioural response, while taking account of uncertainty in the model parameters and heterogeneity in behaviour. The outputs of individual movement models and telemetry data were then used to determine how the probability of exposure might vary among species, individuals, and geographical locations. We then demonstrate how these estimated probabilities of exposure can be translated into probabilities of disturbance. For illustrative purposes, we modelled the potential effects of a temporary decrease in energy assimilation associated with a series of disturbance events that might realistically occur during the construction of an offshore windfarm. Offspring starvation mortality was the vital rate that was most affected by these disturbance events. Monitoring of rate should be considered as standard practice so that populations responses can be detected as early as possible. Predicted effects on individual vital rates depended on the species' movement behaviour and the likely density of animals where the modelled construction activity was assumed to take place. The magnitude of these effects also depended critically on the assumed duration of the reduction in energy assimilation. No direct estimates of this variable are currently available, but we suggest some ways in which it could be estimated. The described approach could be extended to other species and activities, given sufficient information to parameterise the component models. However, we emphasise the need to account for among‐individual heterogeneities and uncertainties in the values of the many model parameters.