This paper analyses the impact of the intensity of improved maize varieties adoption on food security and poverty using data collected in 2010 from maize-legume farming systems in rural Tanzania. We used a continuous treatment approach using generalized propensity score matching and parametric error correction approaches to reduce potential biases stemming from difference in observed characteristics. Estimates of the dose-response functions reveal that average probability of food security, average per capita food expenditure and the average probability of break-even and food surplus increase with the intensity of adoption. On the other hand, the probability of being poor, chronic and transitory food insecurity declines with the intensity of adoption. The results provide strong evidence for heterogeneous food security impacts at different levels of adoption. At low levels of adoption, the average and marginal treatment effects are low while the food security impacts increase substantially at higher level of adoption.