The semiarid region of Brazil consists of a great variety of landscapes, soils and vegetation forms, with complex interrelations. In order to better understand this interplay, we posed two questions: Are there greater pedological similarities among the different landforms of the same catena or among the same landforms from different catenas? Which soil attributes could be the most important to segregate communities of plants? We sampled soils and vegetation on different landforms in four different catenas and performed NMS (non-metric multidimensional scaling) and ANOVA (analysis of variance) to address the first question; also, we carried another NMS following GLM (general linear model regression) to answer the second question. The first NMS indicated the existence of a fertility gradient, grouping communities in relation to similar landforms, confirmed by ANOVA. The second NMS indicated the same gradient whereas the GLM showed that is controlled by aluminum saturation, sodium saturation, phosphorous and sand content. One extreme of the gradient has uplands associated with cerrado vegetation forms whereas the other extreme slopes were associated with dry forests. The lowlands associated with dry forest represent the central position of the fertility gradient. In general, soils at similar landforms showed greater pedological similarity, and their physico-chemical attributes determined the formation and structure of vegetation. This similarity across the same landform refers to the comparable soil formation at each landform and soil age at landscape scale. The characteristics of the vegetation and soils in the Brazilian southern semiarid region indicated a previously wetter climate, during which deep weathered latosols (oxisols) were formed and remain as relics in the present semiarid.