Abstract Background Following infection with Mycobacterium tuberculosis (M.tb) individuals may rapidly develop tuberculosis (TB) disease or enter “latent” infection state with a low risk of progression to disease. The mechanisms underlying this process are incompletely known. Mathematical models use a variety of structures and parameterisations to represent this progression from infection with M.tb to disease. This structural and parametric uncertainty may affect the predicted impact of interventions leading to incorrect conclusions and decision making. Methods We used a simple dynamic transmission model to explore the effect of uncertainty in model structure and parameterisation on the predicted impact of scaling up preventive therapy. We compared three commonly used model structures and used parameter values from two different data sources. Models 1 and 2 are equally consistent with observations of the time from infection to disease. Model 3, produces a worse fit to the data, but is widely used in published modelling studies. We simulated treatment of 5% of all M.tb infected individuals per year in a population of 10,000 and calculated the reduction in TB incidence and number needed to treat to avert one TB case over 10 years. Results The predicted impact of the preventive therapy intervention depended on both the model structure and the parameterisation of that structure. For example, at a baseline annual TB incidence of 500/100,000, the impact ranged from 11% to 27% and the number needed to treat to avert one TB case varied between 38 and 124. The relative importance of structure and parameters varied depending on the baseline incidence of TB. Discussion Our analysis shows that the choice of model structure and the parameterisation can influence the predicted impact of interventions. Modelling studies should consider incorporating structural uncertainty in their analysis. Not doing so may lead to incorrect conclusions on the impact of interventions.