It is well known that the take-up rate of the Dutch housing benefit and othermeans tested benefits is substantially below 100%. In order to measure non-take up oneusually has to simulate entitlement to the benefits. In this paper we take a closer look atthe quality of the simulation. We find evidence that simulation error is much moreimportant than has often been assumed in earlier studies. These studies use the simulationof entitlement in order to select the sample on which an explanatory model for the take-updecision is based. Simulation error may therefore lead to biases in the explanatoryanalysis. Our analysis suggests that a discrepancy between the income that is reported inour database and the income on which the decision to apply for the benefit is based is animportant source of simulation error.The data that were available for this study contain information about refused applicationsand therefore allow us to estimate a richer model than is conventionally used for theanalysis of take-up rates. In this model both the decision of a renting household to applyfor the benefit and that of the authorities to grant or refuse the subsidy are analyzed.Estimation results for this model differ from that of the first.A third model is explicitly based on the assumption that the decision-making processes arebased on an income level that may be different from that reported in the data. Forhouseholds that receive the housing benefit we can compute this alternative income. Thisthird model can be estimated if it is assumed that the same joint distribution of the twoincomes is relevant for households who did not apply for the benefit. Estimation resultsfor this model are again different from those of the other two. They suggest thatmeasurement error in rent, received benefits or housing composition are also more importantthat has been thought.