Multivariate techniques such as factor analysis are common methods for the processing of geodata but they do not take account of the spatial distribution of samples. In his paper, Sandjivy L. (1984) has described the Factorial Kriging Analysis, a new technique based on geostatistics which takes account of the spatial location of data points. He nevertheless notes that the technique is of limited use when there are large number of variables involved. In this case, a complementary method, described in this paper, may be used : Proximity- Analysis. This technique defines a proximity index between samples and optimizes the ratio of the local variance to the total variance as defined on all the variables - by means of the local and total covariance matrix. The method provides linear combinations of variables called proximity factors which describe each variable in terms of spatial variability : the first factors remove background deriving from regionalization, whilst the last ones show only weak spatial variations (low signal to noise ratio) .The interpretation of maps is performed either in terms of "mu1tivariab1e anomalies" or of "multi regional components". The method is applicable to 1D, 2D, or 3D problems with or without discontinuities (limits of basin, independent areas, rock formations, ... ). Two geochemical case-studies are given : a prospection carried out in the Castres region (France) and a stream sediment survey within the Francevillien formations (Gabon).