We present a test of the C(T)2-profile method described by Hill et al. [J. Atmos. Ocean. Technol. 9 (5) (1992) 526] to estimate the surface sensible heat flux over an homogeneous surface. A comparison with traditional eddy correlation measurements performed over a pasture (during the SALSA-Mexico experiment) using three identical large aperture scintillometers (LASs) along a 330 m propagation path and placed at heights 2.50, 3.45 and 6.45 m is first given. Scintillometer derived fluxes using the classical method at one level [Agric. For. Meteorol. 76 (1995) 149] reveal that the three scintillometers provide consistent measurements but underestimate by 15% the flux obtained with the 3D sonic anemometer. This is attributed to spatial non-homogeneities of the experimental site. Considerable scatter (and even the impossibility of performing computations) is found when using the C(T)2-profile method which is particularly prone to errors in nearly neutral and highly unstable conditions. The sensitivity of these errors to the accuracy of scintillometer measurements, the calibration errors and the measurement heights is investigated numerically. Simulations are made assuming a normal distribution of the relative error for C(N)2 with standard deviations σ between 2 and 5% and no calibration error in a first step. Only calibration errors (up to 4% between instruments) are simulated in a second step. They confirm that the profile method degrades very rapidly with the accuracy of C(N)2: for instance the RMS error for H reaches 68 W m-2 (and the cases of impossible computation 28%) for a realistic σ = 5% value, with heights 2.50 and 3.45 m. Results appear slightly less sensitive to small calibration errors. The choice of the measurement heights z1 and z2 is also analysed: a ratio z2/z1 ~ 3 or 4 with z1 > 2 m seems the best compromise to minimise errors in H. Nevertheless the accuracy of the profile method is always much lower than that given by the classical method using measurements at one level, provided a good estimate of roughness length is available. We conclude that the C(T)2-profile method is not suitable for routine applications. (C) 2000 Elsevier Science B.V.