Using representative micro data from the German Socio-Economic Panel Study (SOEP) for the year 2002, we analyse non-take-up behaviour of Social Assistance (SA) in Germany. According to our simulation as much as 67 percent of the eligible population did not claim SA in that year which is slightly higher than reported in previous work. We particularly emphasize the role of measurement error in estimating non-take-up. First, we consider misspecifications of the simulation model due, e.g., to households claiming to have received SA although not simulated as eligible ("beta-error"). Second, we employ sensitivity analyses revealing the impact of measurement errors in reported household income and wealth as well as in simulated needs. Misreported household incomes appear to have the greatest impact on the estimated non-take-up rates, as shown in Monte-Carlo-type simulations. Regression analysis of the potential determinants of non-take-up behaviour confirm that rational motives - i.e., the expected net utility from claiming - as well as stigma and other barriers play a crucial role in explaining the puzzle of large non-take-up rates of SA.