In this paper we aim at studying poverty dynamics and the socio-demographic factors influencing it; alternative measures for the definition of the concept of poverty, making use of panel regression models, are compared. The introduction of two alternative measures is necessary in order to overcome some limitations of the so-called traditional approach. n the traditional approach to poverty measurement a statistical unit (individual or household) is defined as poor if its net equivalent household income is below a certain threshold, the poverty line, ie a percentage of the mean or median of the overall income distribution; let us know consider some limitations. The division of the popularion into the dichotomy of the poor and the non poor seems to be an over simplification; in fact, as pointed out by Cheli and Lemmi (1995) poverty is not a simple attribute that characterises an individual in terms of its presence or absence; we also believe that the relative well-being of a statistical unit is a matter of degree. Moreover, poverty measures based only on monetary variables are not able to capture all the aspects and nuances of the phenomenon; multidimensional indicators seem to be more appropriate in order to describe deprivation better. Finally in a dynamic context, the traditional approach has a further limitation: the mobility of the unitgs near to the poverty line is overestimated. Another aspect to be discussed is that in the context of poverty dynamics analysis there is no unanimity in the choice of the longitudinal units; the controversy is about choosing individuals or households. In any case, it seems reasonable to consider variables regarding the household as poverty indicators, particularly when a multidimensional concept of poverty is considered. On the other hand, it is difficult to define the household as a longitudinal unit of analysis in any rigorous way.