The objective of this paper is to estimate the length of poverty spells and its determinants. We analyze if short-term changes in the labor market affect the probability of staying in poverty. On the assumption that poverty transitions occur more frequently when we use a monthly data rather than an annual data, we are using the panel data of Brazilian Monthly Employment Survey (PME) from March of 2002 to May of 2007. PME is the only panel data which is available in Brazil to carry out this kind of analysis. However, considering that this survey follows households for a very short period, we had to adopt estimation techniques which control cases of right- and left-censoring. The most important results are: how longer the poverty spells, lower the probability to escape from it; households with elderlies have more chances to exit poverty than the others. In the labor market, changes in unemployment rate do not affect directly poverty duration. The effect of unemployment rate occurs indirectly by means of the seasonality of economic activity and of changes in the average wage.