For extrasolar planets discovered using the radial velocity method 1 , the spectral characterization of the host star leads to a mass-estimate of the star and subsequently of the orbiting planet. In contrast, if also the orbital velocity of the planet would be known, the masses of both star and planet could be determined We obtained 51 spectra with a wavelength coverage of 2291 to 2349 nm at a spectral resolution of 100,000. For a detailed description of the observational set-up and data reduction we refer the reader to the supplementary information. During the transit, star-light filters through the atmosphere of the planet, leaving an imprint of molecular absorption lines in the spectrum. In the observed wavelength regime, 56 strong spectral lines from carbon monoxide are expected to be present, and we extracted the CO-signal by cross-correlating a CO model-spectrum with the observed data. We developed our own transmission model to calculate the expected CO-spectrum, with the planet atmosphere described by one mean profile that is in hydrostatic equilibrium (see the supplementary information for details). In our initial model, the temperature is based on the best-fit day-side temperature profile 4 for HD209458b, and gases are uniformly mixed with volume mixing-ratios of 2x10 -4 (CH 4 and CO) and 5x10 -4 (H 2 O) as based on the case 5,6 of HD189733b.The observed spectra are completely dominated by numerous telluric absorption lines, caused mainly by methane and water vapour in the Earth's atmosphere 7 . The depths of these lines vary with airmass, and an important part of the data reduction process therefore involves the removal of this telluric contamination. Residual effects, clearly present at the positions of strong telluric lines, are further suppressed by normalizing each pixel value in a spectrum by its variance over time. Although this may in certain places also suppress carbon monoxide lines, it prevents the cross-correlation signal to be dominated by telluric residuals. Our analysis results in the significant detection of a CO signal and detection of the orbital motion of the planet. Similar crosscorrelation analyses with H 2 O and CH 4 templates did not result in detections. The model transmission spectrum of CH 4 is so densely packed with lines that it strongly hampers a cross-correlation analysis. Two CH 4 lines are significantly stronger than the others, but unfortunately one falls in a gap between two of the detector arrays. Cross-correlating with a model spectrum combining CO with H 2 O and CH 4 does not give an improvement over the signal from CO alone. The result of the CO cross-correlation is shown in Figure 1. Due to the orbital velocity of the planet, of which the radial component changes during the transit, the CO-signal shifts in position from the beginning to the end of the transit by ~30 km sec -1 .As shown in Figure 2, from this shift we derive the planet orbital velocity to be v p = 140 ± 10 km sec -1 (1σ), corresponding to the maximum radial velocity at quadrature. In combination wit...