We apply dynamic probit models allowing for unobserved heterogeneity and endogenous initial conditions to IT-SILC data to investigate the low income persistence of households with disabled members. We find that their probability of being in a low income state is higher when compared with households without disabled members. In both cases household head’s characteristics, as employment status and education, contribute to determine low income positions. Our results also support the hypothesis of endogenous initial conditions. Both unobserved heterogeneity and state dependence are important to determine low income positions. Our findings suggest that a structural intervention geared at lifting households out of low income in future requires to get them out of low income at present. Moreover, preventing rather than rescuing actions are preferable.