The IFI's rely on twin propositions that poverty alleviation is best pursued through increased growth, and that trade liberalisation encourages growth and thereby poverty alleviation. These aggregate propositions are not disputed in this paper. Rather, the argument is made for disaggregation to identify the winners and losers among the poor in the short and medium run from further trade policy liberalisation, both between and within countries. Disaggregation is important on both equity grounds, especially when the losers are among the poor. It is also important on efficiency grounds particularly when designing policies to help poor losers realise the opportunities for gain from trade policy liberalisation in the longer run. Two methodologies are frequently employed to assess the linkages between trade and poverty. Country and sector case studies dominate the literature. The key difficulty with case studies is that it is not possible to deploy their rich descriptive data in a consistent analytical framework. It is usually not possible to construct a quantitative counterfactual situation, for example the impact of a trade policy change on the poor. Obtaining a counterfactual through general equilibrium trade models has its own catalogue of difficulties arising from data availability, model assumptions and interpretation of results. The two methodologies lie at extreme ends of a spectrum, but insights from both can be mutually reinforcing. This paper uses Zambia as an 'example' country to explore the possibility of combining the rich poverty case study material available for that country with the results for Zambia of a multilateral trade model based on the GTAP dataset and modelling software. It describes some of the salient features of case studies of poverty in Zambia. A common thread runs through all the case studies, that trade policy reform in Zambia is likely to be pro-poor. This key proposition is tested using the GTAP dataset and modelling software. The GTAP database for 1997 is described including an extension for Zambia to permit the analysis of poverty impacts for four classes of households together with the modelling strategy adopted for this exploratory study. It then reports on the poverty impacts of a series of trade policy experiments using the 1996 LCMS Survey for Zambia to estimate headcount changes from: • Unilateral trade policy reforms in Southern Africa that took place from 1992-4 up to 1997. • A seven-country version of the SADC FTA. • A 'suppose' WTO Round. • A 'suppose' effective extension of the EU/South Africa FTA into an EU/SADC7 FTA through Least Developed Country access into the EU through negotiations about to begin. Zambian households are disaggregated into four groups so that GTAP aggregate household results can be disaggregated to real post-tax income changes in each of the four household groups in post simulation calculations. A key finding is that regionally ii based trade policy reforms have a neutral or adverse impact on household income distribution compared with possible major trade policy reforms under the WTO. However, the final headcount poverty impacts of the international trade policy reforms are offset by the lower income responsiveness of the poverty impacts in the poorest rural households. Whilst these findings are suggestive, the research strategy on the modelling side using both the GTAP dataset and runGTAP as computing software has a number of important limitations. On the data side, for Southern Africa applications, the standardised GTAP dataset throws away too much useful information for trade and poverty analysis that is available in the underlying MERRISA SAMs upon which the Southern Africa dataset for GTAP was built. This is most notable for household aggregation, factor aggregation at the low income or subsistence end, margins aggregation, rudimentary treatment of government income and expenditure. On the runGTAP modelling side, satisfactory resolution of the database problems poses serious programming problems. The above suggests an alternative trade and poverty research strategy in which the CGE model and country case study interface is first explored with country models without the constraints of the standard GTAP dataset aggregation. An obvious choice for a starting model is the standard model Loefgren, Harris and Robinson (2001). At a later point, such country models could be tied into a global model using the GTAP dataset for scenario calculations using a common sectoral classification to complete the bottom-up strategy for the analysis of trade and poverty impacts.