Chain drift is the difference between the rate of change calculated by chaining an index over a multi period interval and that obtained using endpoints only. This intransitivity can produce ambiguity in estimated growth rates if the “correct” linking interval is not known. The present paper compares the propensity for chain drift in two leading superlative index number formulas: the Fisher and the Tornqvist. It is shown, both empirically and theoretically, that the Tornqvist index is less affected by chain drift than is the Fisher. Under certain conditions, the sign of chain drift in the Fisher index can be predicted.