The use of thermal IR for estimating regional evaporation has been tested on the Crau plain (50 km x 30 km) northwest of Marseille. Within it, 4 sufficiently homogeneous units were selected, each large enough for use of satellite remote sensing (from 50 to 200 km2). One corresponds to the dry Crau, while the three others are located within the irrigated Crau. Two sites (dry Crau and one irrigated) were chosen for ground test studies, corresponding to microclimate and energy budget measurements continuously recorded during 3 to 4 months in the summers of 1978 and 1979. The data obtained first allow the characterization of microclimatic differences between dry and irrigated Crau, either within a given day (fig. 4), or on a long-term basis (fig. 5) where surface temperature differences of up to 20-25° may appear at midday. The main interest of these ground data is to test the feasibility of using Ts measurements for estimating evaporation. When based on the currently adopted energy budget method combined with aerodynamic transfer, this appears to work well for the irrigated area. But, in dry conditions, it greatly depends upon the estimate of roughness parameter Zo, and the adoption of the THOM correction (fig. 7). On a long-term basis, the simplified approach proposed by JACKSON et al. (1977) ET = A + Rn - B (Ts - Ta), using one single surface temperature data, appears as satisfactory as the complete method. But one has to adopt local parameters A and B, fairly different from the values originally proposed by the authors (fig. 8). The reasons for this are analysed and discussed. Finally, comparisons with single approaches ( formula, see attached document ) give no definitive superiority for surface temperature-based methods, since it seems to be really difficult to ensure precision better than 10 to 20 %. On the other hand, these last methods work equally for the two sites and give the correct differences between dry and irrigated Crau. The single methods, however, are only appropriate to one or the other of these conditions (fig. 9, 10, 11, 12). Use of thermal I.R. for estimating evaporation thus appears to be problematic for fixing absolute values (unless locally adapted by using known ground parameters), but really promising for evaluating quantitative spatial variations.