Spatial relationships between tree species and hydrological soil constraints are analysed within a 10-ha rainforest plot at Piste de St Elie in French Guiana. We used canonical correlation analysis to cross directly the occurrence-by- species table of 4 992 individuals (d.b.h. $>$ 10 cm) belonging to 120 species with qualitative soil variables and quantitative spatial data. Firstly, the list of species occurrences was confronted to nine soil descriptors characterising a weathering sequence from the initial well-drained ferralitic cover to transformed hydromorphic soil conditions. This analysis revealed that, apart from some specialised species restricted to the swamps that experience prolonged water saturation, the most abundant species can be ordered along two intermingled gradients of tolerance limiting their niche amplitude: a main gradient of tolerance to prolonged water saturation that appears down slope during the weathering sequence; a second gradient of less importance, displaying the species intolerant of prolonged water saturation according to their tolerance to temporary confinement of the uphill transformed soil systems due to the late appearance of a perched water-table. The results support the hypothesis that at Piste de St Elie, the constraining soil conditions imposed by surface water saturation are more important determinants for tree zonation of many tree species than water shortage. Secondly, the list of species occurrences was confronted to a spatial data table built from a trend surface regression of the tree coordinates. This analysis indicated that soil drainage is the main structuring factor of the local multispecies spatial pattern. After partialling out the soil effect, the multispecies pattern revealed a broader scale of heterogeneity that we supposed to be linked to endogenous factors resulting from population dynamics. Implications of the results are then discussed in the perspective of future research on tree zonation, local diversity pattern and community structuring in tropical rainforests.