The geochemistry of multiply substituted isotopologues (‘clumped-isotope’ geochemistry) examines the abundances in natural materials of molecules, formula units or moieties that contain more than one rare isotope (e.g. 13C18O16O, 18O18O, 15N2, 13C18O16O22−). Such species form the basis of carbonate clumped-isotope thermometry and undergo distinctive fractionations during a variety of natural processes, but initial reports have provided few details of their analysis. In this study, we present detailed data and arguments regarding the theoretical and practical limits of precision, methods of standardization, instrument linearity and related issues for clumped-isotope analysis by dual-inlet gas-source isotope ratio mass spectrometry (IRMS). We demonstrate long-term stability and subtenth per mil precision in 47/44 ratios for counting systems consisting of a Faraday cup registered through a 1012 Ω resistor on three Thermo-Finnigan 253 IRMS systems. Based on the analyses of heated CO2 gases, which have a stochastic distribution of isotopes among possible isotopologues, we document and correct for (1) isotopic exchange among analyte CO2 molecules and (2) subtle nonlinearity in the relationship between actual and measured 47/44 ratios. External precisions of ∼0.01‰ are routinely achieved for measurements of the mass-47 anomaly (a measure mostly of the abundance anomaly of 13C-18O bonds) and follow counting statistics. The present technical limit to precision intrinsic to our methods and instrumentation is ∼5 parts per million (ppm), whereas precisions of measurements of heterogeneous natural materials are more typically ∼10 ppm (both 1 s.e.). These correspond to errors in carbonate clumped-isotope thermometry of ±1.2 °C and ±2.4 °C, respectively. Copyright © 2009 John Wiley & Sons, Ltd.