Summary: This paper derives optimal inheritance tax formulas that capture the key equity-efficiency trade-off, are expressed in terms of estimable sufficient statistics, and are robust to the underlying structure of preferences. We consider dynamic stochastic models with general and heterogeneous bequest tastes and labor productivities. We limit ourselves to simple but realistic linear or two-bracket tax structures to obtain tractable formulas. We show that long-run optimal inheritance tax rates can always be expressed in terms of aggregate earnings and bequest elasticities with respect to tax rates, distributional parameters, and social preferences for redistribution. Those results carry over with tractable modifications to (a) the case with social discounting (instead of steady-state welfare maximization), (b) the case with partly accidental bequests, (c) the standard Barro-Becker dynastic model. The optimal tax rate is positive and quantitatively large if the elasticity of bequests to the tax rate is low, bequest concentration is high, and society cares mostly about those receiving little inheritance. We propose a calibration using micro-data for France and the United States. We find that, for realistic parameters, the optimal inheritance tax rate might be as large as 50--60\% -- or even higher for top bequests, in line with historical experience.