A unique methodology to measure gas fluxes of SO2 and NO2 from ships using optical remote sensing is described and demonstrated in a feasibility study. The measurement system is based on Differential Optical Absorption Spectroscopy using reflected skylight from the water surface as light source. A grating spectrometer records spectra around 311 nm and 440 nm, respectively, with the telescope pointed downward at a 30A degrees angle from the horizon. The mass column values of SO2 and NO2 are retrieved from each spectrum and integrated across the plume. A simple geometric approximation is used to calculate the optical path. To obtain the total emission in kg h(-1) the resulting total mass across the plume is multiplied with the apparent wind, i.e. a dilution factor corresponding to the vector between the wind and the ship speed. The system was tested in two feasibility studies in the Baltic Sea and Kattegat, from a CASA-212 airplane in 2008 and in the North Sea outside Rotterdam from a Dauphin helicopter in an EU campaign in 2009. In the Baltic Sea the average SO2 emission out of 22 ships was (54 +/- 13) kg h(-1), and the average NO2 emission was (33 +/- 8) kg h(-1), out of 13 ships. In the North Sea the average SO2 emission out of 21 ships was (42 +/- 11) kg h(-1), NO2 was not measured here. The detection limit of the system made it possible to detect SO2 in the ship plumes in 60% of the measurements when the described method was used. A comparison exercise was carried out by conducting airborne optical measurements on a passenger ferry in parallel with onboard measurements. The comparison shows agreement of (-30 +/- 14)% and (-41 +/- 11)%, respectively, for two days, with equal measurement precision of about 20%. This gives an idea of the measurement uncertainty caused by errors in the simple geometric approximation for the optical light path neglecting scattering of the light in ocean waves and direct and multiple scattering in the exhaust plume under various conditions. A tentative error budget indicates uncertainties within 30-45% but for a reliable error analysis the optical light path needs to be modelled. A ship emission model, FMI-STEAM, has been compared to the optical measurements showing an 18% overestimation and a correlation coefficient (R-2) of 0.6. It is shown that a combination of the optical method with modelled power consumption can estimate the sulphur fuel content within 40%, which would be sufficient to detect the difference between ships running at 1% and at 0.1%, limits applicable within the IMO regulated areas.