On the basis of a longitudinal administrative dataset(1991-2002) merged with the Census of 2001 and the National Register, the majority of the poor elderly in Belgium appear to be persistently poor. The question arises why this might be so. To the extent that individual characteristics such as low abilities persist over time, they may also be the reason that individuals persist in poverty over time. In that case, one expects that once individual characteristics are controlled for, duration dependence in poverty becomes spurious. The alternative possibility is that poverty experience has a causal impact on future poverty. This may be because of a poverty trap: people may be given an incentive not to work while at the same time they slip into poverty. Or this may be due to depreciation of human capital or loss of motivation. The reasons for dependence in poverty are of interest for developing effective poverty reducing measures since true dependence would suggest to focus on stigma and adverse work incentives while spurious dependence would suggest to change individual’s characteristics. The simultaneous estimation of a multiple-spell discrete-time hazard model of transitions in and out of poverty, that allows for unobserved effects and a significant initial condition problem, lends strong empirical support for true duration dependence in poverty. This suggestion sounds reasonable since in Belgium elderly unemployed are exempted from the search for a job and thus easily exposed to depreciation of human capital and employers are reluctant to invest in the human capital of older workers. In addition in Belgium both employers and the government design retirement pathways that give elderly strong incentives to leave the labour market as soon as possible.