This paper tries to identify the correlates of poverty in urban Eritrea using an estimation technique (the DOGEV model) that also allows for the inclusion of a measure of a??persistencea?? in poverty levels from cross-sectional estimation. The results suggest that 17 percent of the probability of being moderately poor and 22 percent of the probability of being extremely poor in Eritrea was attributable to this a??persistencea??a??a predisposition toward poverty likely due to latent attributes related to past experience of poverty itself. The results also suggest that, in the post-war economy of the mid-1990s, those with vocational training fared best among all education groups. Being a war veteran also had a strong negative association with the povertya??reflecting successful attempts to support that group. The receipt of remittances also reduced the likelihood of poverty; though receipts from outside Eritrea had a much stronger effect than receipts from within Eritrea.