Defining urban morphology in terms of the shape and density of urban land use has hitherto depended upon the informed yet subjective recognition of patterns consistent with spatial theory. In this paper we exploit the potential of urban image analysis from remotely sensed data to detect, then measure, various elements of urban form and its land use, thus providing a basis for consistent definition and thence comparison. First, we introduce methods for classifying urban areas and individual land uses from remotely sensed images by using conventional maximum likelihood discriminators which utilize the spectral densities associated with different elements of the image. As a benchmark to our classifications, we use smoothed UK Population Census data. From the analysis we then extract various definitions of the urban area and its distinct land uses which we represent in terms of binary surfaces arrayed on fine grids with resolutions of approximately 20 m and 30 m. These images form surfaces which reveal both the shape of land use and its density in terms of the amount of urban space filled, and these provide the data for subsequent density analysis. This analysis is based upon fractal theory in which densities of occupancy at different distances from fixed points are modeled by means of power functions. We illustrate this for land use in Bristol, England, extracted from Landsat TM-5 and SPOT HRV images and dimensioned from population census data for 1981 and 1991. We provide for the first time, not only fractal measurements of the density of different land uses but measures of the temporal change in these densities.