A two-phase estimation of coverages of the k land categories partitioning a study area is considered. In the first phase a sample of N points is selected according to a spatial sampling design and the selected points are assigned to the k categories on the basis of satellite imagery or aerial photos. Simple random sampling, stratified sampling and systematic sampling of points are considered. The theoretical results suggest the use of stratified sampling as the best design-based spatial strategy in the first phase. Subsequently, in the second phase, a sample of n < N points is selected using a finite population sampling design. Then the selected points are visited and correctly classified on the ground. To this purpose, a stratified sampling is considered in which the strata are determined by the aerial classification. The statistical properties of the resulting two-phase estimator are derived and a conservative estimator of its variance–covariance matrix is proposed. The procedure is checked on the basis of a simulation study performed on some artificial populations. Copyright © 2004 John Wiley & Sons, Ltd.