The methods for analyzing data in one dimension (usually time) are highly developed. However, in several dimensions, most applied workers only use one class of methods -- that in which the data is assumed to be the sum of a deterministic trend and an uncorrelated random error. The more general model with a spatially correlated error has been neglected in most fields -- oceanography is a conspicuous exception. This neglect is partly due to the lack of expositions for the applied workers. However, the manner in which much geological data is now collected does make this application difficult. The aim of this paper is to explain the relevance of the possible models and methods and to indicate their data requirements. An appendix has been added on Matheron's work.