The probit model is widely used to analyse take-up behaviour by people eligible for social security benefits. In practice, observed income and simulated benefit entitlement are error-prone, leading to incorrect identification of eligible individuals and measurement error in the explanatory variables. We analyse the consequences in two cases: a simple model where takeup is perfectly observed but measured income is unreliable and a more complex model where both income and benefit receipt are error-prone. We find that applied results may be very sensitive to measurement error; that biases are qualitatively complicated and that small-ó approximations perform poorly.