This paper addresses the question: What kind of households are vulnerable and how are they vulnerable in Pakistan? This question is investigated using two-period panel data (surveyed in 2001 and 2004) covering about 1,600 households in rural Punjab and Sindh, and four rounds of nationally-representative, repeated cross-section data, covering about 15,000 households in each round of 1998/99, 2001/02, 2004/05, and 2005/06. During this period, average consumption initially decreased and then increased. Associated with this change, poverty increased initially and then decreased, and inequality decreased initially and then increased. The vulnerability analysis in this paper focuses on the second period when poverty decreased. Five measures of vulnerability are employed: transient poverty components of observed poverty, decreases in consumption levels, sensitivity of consumption changes to village-level shocks, variance of consumption changes, and the welfare cost of risk simulated under the assumption of a specific utility function. Empirical results are summarized as follows. Important physical assets in Pakistan, i.e., farmland, livestock, and durable goods, are vulnerablity-reducing in general. The landed households, however, may have difficulty in catching up with the macroeconomic growth rate in a boom. Access to non-farm employment is vulnerablity-reducing. In contrast, access to credit and remittance has mixed effects, probably due to the reverse causality that households hit by adverse shocks seek credit or remittance more eagerly. Education is weakly associated with higher vulnerability. This could be because the welfare level of educated households is higher than uneducated households in general, implying that educated households have larger room for consumption curtailment when hit by an adverse shock. Households with more dependent members are less vulnerable, suggesting the existence of an informal social support or implicit contract for households with more children. Larger households suffer from a larger welfare cost of risk than smaller households do. Geographically, residents in rural Sindh are more subject to various types of vulnerability than those in rural Punjab, especially northern districts of Punjab. Across the country, however, residents in NWFP and Balochistan suffer a larger cost of welfare loss due to risk, making the difference between rural Sindh and rural Punjab a minor one, and urban residents in Punjab and Sindh are less subject to vulnerability than all others, although we have to be careful since the regional contrasts in vulnerability across Pakistan are not based on panel data. To estimate the welfare cost of risk from repeated cross-section data, we impose restrictions that correspond to the permanent income hypothesis with perfect credit markets, but the dynamics of consumption inequality is not wholly consistent with these restrictions.