Design-based estimators of totals, means or proportions in finite populations generally are functions of weighted sums. If each element selected into the sample is also observed, then for the calculation of the pi-estimator these weights are just the inverse inclusion probabilities of the elements. However, if e.g. nonresponse or attrition over time occurs, calculation of these weights also includes modeling of nonresponse and/or attrition mechanisms. Since models of these mechanisms are disputable, `pure' design weights can be the basis for cal- culating alternative weights by a different modeling e.g. of nonresponse and/or attrition mechanisms. In the case of complex sampling schemes, however, it is often not possible to derive the exact inclusion probabilities. In that case, weights may be derived based on approximations and/or simplifying assumptions. In this paper, after describing the selection schemes of the subsamples A, B, C, D and E of the German Socio-Economic Panel (GSOEP), approximate design weights are derived which enable users of the GSOEP to calculate their own weights if desired.