Contemporary preclinical in vitro and in vivo methods have been imperfect in predicting drug-induced Torsades de Pointes (TdP) in humans. A better understanding of additional relevant factors in the genesis of drug-induced TdP is necessary. New sophisticated in vitro techniques, such as arterially perfused ventricular wedge preparations or isolated perfused hearts, potentially offer a better understanding of torsadogenic mechanisms and a refinement of drug testing. Of particular interest are the dispersion of repolarization and the refractoriness of different cell types across the ventricular wall, triangulation of the action potential, reverse use dependence and instability of the action potential duration. In vivo models are currently refined by establishing parameters such as beat-to-beat variability and T-wave morphology as derived from the in vitro proarrhythmia indices. Animal models of proarrhythmia are to date not recommended for routine evaluation. A pharmacodynamic interaction with combinations of torsadogenic compounds is another area to be considered. Little is known about channel/receptor cross talk, although considerable evidence exists that cardiac G protein-coupled receptors can modulate hERG channel function. More investigations are necessary to further evaluate the role of altered gene expression, mutations, and polymorphisms in drug-induced TdP. A novel mechanism of drug-induced torsadogenesis is the reduced expression of hERG channel protein on the plasma membrane due to a trafficking defect. Pharmacokinetic and metabolism data are crucial for calculating the risk of a torsadogenic potential in man. Consideration of intracardiac accumulation can help in delineating pharmacokinetic-pharmacodyamic relationships. In silico virtual screening procedures with new chemical entities to predict hERG block may develop as a promising tool. The role of in silico modeling of TdP arrhythmia is likely to become increasingly important for organizing and integrating the vast amount of generated data. At present, however, in silico methods cannot replace existing preclinical in vitro and in vivo models.