The retrieval of surface parameters from remotely sensed data is of prime interest for the estimation of surface properties of various planets in the solar system, including the Earth. Bidirectional reflectance measurements taken over natural surfaces in visible and near-infrared spectral bands represent one data set from which these surface properties could be estimated. To achieve this goal, it is necessary to have both physical models predicting the bidirectional reflectance field as a function of the relevant surface parameters, and numerical procedures allowing the inversion of these models using a limited sampling of the bidirectional reflectance field. Given that theoretical models of the bidirectional reflectance have been published (Hapke, 1981, 1984, 1986; Verstraete et al., 1990) and that numerical procedures are available (Pinty et al., 1990), this paper focuses on the errors and uncertainties in the retrieved parameters which may arise because of (1) the weaknesses in our theoretical understanding and representation of the surface radiation transfer and (2) the errors in the bidirectional reflectance data. For instance, it is shown that the addition a posteriori of an amplitude parameter in the function accounting for the opposition effect can drastically modify the retrieved values of the optical and morphological parameters of the surface. The consequences of uncertainties in the reflectance data are also investigated, and the redistribution, by the inversion procedure, of such uncertainties on the retrieved parameters is discussed. Finally, synthetic reflectance data contaminated by a known random noise are used to examine the numerical stability of the retrieval and the compatibility between the models. ¿ 1991 American Geophysical Union