The potential of the LAS (large aperture scintillometry) method for measuring sensible heat flux (H) directly integrated over a two-field composite surface is evaluated. We describe a field experiment performed within the Alpilles/ReSeDa project in the south-east of France over a composite surface made up of wheat and bare soil (451 and 216 m long respectively) using two 0.15-m diameter scintillometers mounted at heights of 2.05 and 4.54 m. When compared against reference values obtained by the eddy correlation technique, LAS-measured sensible heat flux reveals a systematic overestimation of about 10%. A simple model describing the integration of the scintillometer signal along the beam for a two-field composite surface is described. A simulation of the experiment confirms that the bais observed is related to non-linearities in the integration process in relation with the bell-shape sensitivity curve of the instrument to the structure parameter for the refractive index it measures. The model is used to test the sensitivity of the LAS-derived H values to the composition of the pathlength (ratio of both surfaces) and to the contrast in sensible heat flux and roughness length between the two fields. Sensitivity tests to the aggregation scheme for roughness length (two of them are tested) and to the measurement height are also presented. The composition of the surface in combination with the contrast in sensible heat flux (in direct relation with the contrast in latent heat flux) explains most of the bias, with possible deviations ranging from -50 up to 80 W m-2. A tentative semi-empirical method for correcting the bias is suggested, which only requires a crude estimate of the contrast in component sensible heat fluxes along the pathlength.