Measurements of the surface albedo of Haut Glacier d'Arolla (Switzerland) derived from Landsat Thematic Mapper data (alphaTM) are compared with ground-based albedo measurements (alphag). The analysis is based on (a) 137 aground-based measurements made at 67 sites during three extensive glacier surveys in the 1993 ablation season, and (b) data acquired during three corresponding cloud-free Landsat overpasses on 29 May, 1 and 17 August. For the first overpass large differences in the snow albedo are found (average:alphaTM 0.73, alphag0.55), which are attributed to snowfall and metamorphism of snow between the times of the satellite overpass and the ground measurements. For the second overpass there is much better agreement between the two snow albedos (average: alphaTM 0.56, alphag 0.51). For the second and third overpass there is good agreement between alphaTM and alphag of glacier ice (average:alphaTM 0.17, alphag0.19). However, the scatter in alphaTMalpha g is large which is very probably related to the high spatial variability of the ice albedo over distances of only a few metres and the smallscale coverage (in terms of ground area) of the ground measurements compared to the TM pixel size. In addition, there is a tendency for alphaTM to overestimate low values of alphag and to underestimate high values. This is explained by a possible bias in the ground-based measurements towards selecting either relatively clean or relatively dirty ice surfaces. Several errors associated with uncertainties in the method of albedo retrieval are discussed. It is suggested that the assumption of an isotropically reflecting surface is a main source of error in the satellite-derived albedo.