The uncertainty in the Standard-Model prediction for the anomalous magnetic moment of the muon is dominated by strong interaction contributions. While the accuracy of hadronic vacuum polarization can be improved by dedicated e^+e^- --> hadrons cross-section measurements, the evaluation of hadronic light-by-light scattering currently relies on models and is likely to constitute the road block for fully exploiting the upcoming g-2 experiments at FNAL and J-PARC. In the talk I will review our recently developed dispersive formalism that allows for a data-driven analysis of the most important hadronic channels in the light-by-light tensor, and discuss the experimental ingredients required for the evaluation of one- and two-pion intermediate states.