Torsade de Pointes (TdP) is a well-described major risk associated with various kinds of drugs. However, prediction of this risk is still uncertain both in preclinical and clinical trials. We tested 45 reference compounds on the model of isolated canine Purkinje fibres. Of them, 22 are clearly associated and/or labelled with a risk of TdP, and 13 others are drugs with published clinical evidence of QT prolongation, with only one or two exceptional cases of TdP. The 10 remaining drugs are without reports of TdP and QT prolongation. The relevance of different indicators such as APD(90) increase, reverse use dependency, action potential triangulation or effect on V(max) was evaluated by comparison with available clinical data. Finally, a complex algorithm called TDPscreen and based on two subalgorithms corresponding to particular electrophysiological patterns was defined. This latter algorithm enabled a clear separation of drugs into three groups: (A) drugs with numerous or several reports (>2 cases) of TdP, (B) drugs causing QT prolongation and/or TdP only, the latter at a very low frequency (< or =2 cases), (C) drugs without reports of TdP or QT prolongation. The use of such an algorithm combined with a database accrued from reference compounds with available clinical data is suggested as a basis for testing new candidate drugs in the early stages of development for proarrhythmic risk prediction.