Photoclinometry, the use of brightness variations to indicate the tilt from the horizontal of the surface elements represented by the pixels of an image, is discussed. Three photoclinometry techniques are being studied: (a) the discrimination of significantly different modes in brightness histograms, which lead to spatially correlated regions when mapped back into image space; (b) the use of multispectral information about a scene to classify the pixels into a (small) number of spatially related units which are assumed in general to have different albedos; and (c) the use of self-adaptive, two-dimensional time-series-analysis methods to locate boundaries between regions of differing albedo. Currently, method (a) is regarded as least reliable, whereas method (c) is potentially applicable to a wide range of image classification problems. Tests of the ability of the photoclinometric method to reproduce known topography show that at a spatial resolution (pixel size) of order 100 m, it is capable of estimating absolute topographic heights to within 10% over horizontal distances of order 100 km.