Abstract. The Orbiting Carbon Observatory-2 (OCO-2), launched on 2 July 2014, is a NASA mission designed to measure the column-averaged CO₂ dry air mole fraction, X_CO2. Towards that goal, it will collect spectra of reflected sunlight in narrow spectral ranges centered at 0.76, 1.6 and 2.0 μm with a resolving power (λ/Δ λ) of 20 000. These spectra will be used in an optimal estimation framework to retrieve X_CO2. About 100 000 cloud free soundings of X_CO2 each day will allow estimates of net CO₂ fluxes on regional to continental scales to be determined. Here, we evaluate the OCO-2 spectrometer performance using pre-launch data acquired during instrument thermal vacuum tests in April 2012. A heliostat and a diffuser plate were used to feed direct sunlight into the OCO-2 instrument and spectra were recorded. These spectra were compared to those collected concurrently from a nearby high-resolution Fourier Transform Spectrometer that was part of the Total Carbon Column Observing Network (TCCON). Using the launch-ready OCO-2 calibration and spectroscopic parameters, we performed total column scaling fits to all spectral bands and compared these to TCCON results. On 20 April, we detected a CO₂ plume from the Los Angeles basin at the JPL site with strongly enhanced short-term variability on the order of 1% (3–4 ppm). We also found good (< 0.5 ppm) inter-footprint consistency in retrieved X_CO2. The variations in spectral fitting residuals are consistent with signal-to-noise estimates from instrument calibration, while average residuals are systematic and mostly attributable to remaining errors in our knowledge of the CO₂ and O₂ spectroscopic parameters. A few remaining inconsistencies observed during the tests may be attributable to the specific instrument setup on the ground and will be re-evaluated with in-orbit data.