Rationale: Determination of δ17O values directly from CO2 with traditional gas source isotope ratio mass spectrometry is not possible due to isobaric interference of 13C16O16O on 12C17O16O. The methods developed so far use either chemical conversion or isotope equilibration to determine the δ17O value of CO2. In addition, δ13C measurements require correction for the interference from 12C17O16O on 13C16O16O since it is not possible to resolve the two isotopologues. Methods: We present a technique to determine the δ17O, δ18O and δ13C values of CO2 from the fragment ions that are formed upon electron ionization in the ion source of the Thermo Scientific 253 Ultra high-resolution isotope ratio mass spectrometer (hereafter 253 Ultra). The new technique is compared with the CO2-O2 exchange method and the 17O-correction algorithm for δ17O and δ13C values, respectively. Results: The scale contractions for δ13C and δ18O values are slightly larger for fragment ion measurements than for molecular ion measurements. The δ17O and Δ17O values of CO2 can be measured on the 17O+ fragment with an internal error that is a factor 1–2 above the counting statistics limit. The ultimate precision depends on the signal intensity and on the total time that the 17O+ beam is monitored; a precision of 14 ppm (parts per million) (standard error of the mean) was achieved in 20 hours at the University of Göttingen. The Δ17O measurements with the O-fragment method agree with the CO2-O2 exchange method over a range of Δ17O values of −0.3 to +0.7‰. Conclusions: Isotope measurements on atom fragment ions of CO2 can be used as an alternative method to determine the carbon and oxygen isotopic composition of CO2 without chemical processing or corrections for mass interferences.