Based on a large-scale household survey conducted in low-income neighborhoods in six large Chinese cities, this paper applies multi-measurement and hierarchical linear regression models to present different poverty profile in different cities and to examine the determinants of poverty incidence. Referring to the theory of welfare triangle proposed by Rose 1986), this paper takes the influence of household, market, and state into account to examine the mechanism of poverty incidence in Chinese cities. The empirical results reveal salient intra-urban differences in terms of the variation of poverty intensity and composition among the six cities. Guangzhou and Xi'an suffer from the severe problem of relative deprivation and their low-income groups have very complex composition. The hierarchical linear regression models demonstrate that urban poverty incidence is contingent upon variations at individual, neighborhood, and city levels. In general, neighborhood with better economic status reduces the individual's risk of urban pauperization, while neighborhood with strong poverty culture or long poverty history tends to increase the risk of pauperization.