Correspondence analysis is a dual Q and R mode factor analysis that allows simultaneous visualization of relationships between normalized variables and samples. The usefulness of the technique is first illustrated with an analysis of a simulated feldspar compositional data set and then applied to analysis of patterns among variables and samples for compositions obtained for Viking Lander 1 soils and Pathfinder rocks and soils. For the Martian data the first two factors capture 90% of the variance. Data projected onto the plane defined by these two factors fall within a mixing triangle defined by three geologically plausible end-members: basaltic shergottite, andesite, and a halide-kieserite evaporite. However, examination of data projected onto the second and third factors shows that the compositions are systematically displaced from the mixing triangle plane in the same general direction as the Fe vector. Inclusion of a Fe oxide (e.g., hematite or maghemite) as a reasonable fourth end-member forms a mixing tetrahedron that encapsulates the data. Least squares mixing analysis of the data with these four end-members and the elemental suite measured at both landing sites (Si, Al , Ti, Fe, Mg, Ca, S, Cl) implies that aeolian dust on rocks is a physical mixture of basalt and andesite-derived materials in approximately equal proportions, soil components at the two landing sites are similar to each other and less basalt-like than the aeolian dust, and Viking 1 soils are more andesite-rich than the Pathfinder soils. ¿ 2000 American Geophysical Union