We present a new method of retrieving aerosol optical thickness over land that takes advantage of the dual look capability of the ATSR-2 instrument. Unlike some methods, the approach does not require dense dark vegetation to be located, nor is it dependent upon invariant targets, nor does it assume that the underlying surface reflectance is purely Lambertian. It makes use of a new surface approximation that exploits the property of the surface which is least variable with wavelength to allow atmospheric parameters to be retrieved and, consequently, is applicable over virtually any land surface. The method is tested using simulated ATSR-2 measurements based upon a radiative transfer model coupled with a surface bidirectional model that includes the hot-spot effect, and is shown to be robust to the effects of differing underlying canopy types. At high optical depths, the retrieval process achieves a mean error in optical thickness of less than 2.5%, depending on surface type, with a maximum absolute error (for the same data sets) of 5% while at low optical depths errors are of the order of 5%. The effects of systematic error in the simulated radiances are also shown. The method may be readily extended to other multiple look satellite radiometers.