Abstract. One hundred and sixty-six soil samples representing eleven textural classes and having wide variations in organic matter (OM) contents and other physicochemical properties were collected from different locations in southeastem Nigeria to study the relationship between mechanical composition and saturation water percentage (SP). The objective was to develop a prediction model for silt+clay (SC) and clay (C) contents of these soils using their SP values. The magnitude of the correlation coeffcients (r) between SC or C and SP was depend­ent on the amount of organic matter (OM) present in the soils. For soils with <1.00 % OM, the correlation (r) be­tween SC and SP was 0.9659 (p <0.001) and that between C and SP was 0.9539 (p < 0.001). For soils with > 2.00 % OM, the ‘r’ values were generally Iow, varying between 0.5320 and 0.2665 for SC and 0.6008 and 0.3000 for C. The best-fit regression models for predicting SC and C were developed with soils having <1.00 % OM. An inde­pendent data set from 25 soil samples collected from other parts of the study area was used to test the predictive ability of the best-fit models. These models predicted SC and C accurately in soils having between 0.28 and 1.10 % OM, but poorly in soils having between 1.31 and 3.91 % OM. These results show that the use of saturation water percentage to predict the mechanical composition of soils is most reliable for soils with low (<1.00 %) OM contents.