The determination of water retention characteristics of soils in tropical regions is hampered by the lack of adequate laboratory equipment and high cost of analysis. In this paper the possibilities of estimating water retention characteristics from routinely determined properties of Ferralsols/Oxisols and related soils are investigated. Two sets of data on soils from South America, Africa and South East Asia, were analyzed. Set 1 (91 samples from 31 profiles) was used for multiple linear regression to correlate water retention and parameters of the ‘Van Genuchten equation’ (a semi-empirical expression for the water retention curve) to other soil properties. Set 2 (35 samples from 13 profiles) was used as an independent data set to test the results obtained for Set 1. In Set 1, clay and silt accounted for 84% of the variance in moisture content at 10 kPa (θ10 kPa) and for 80% at 1.5 MPa (θ1.5 MPa). Adding organic carbon to the regression equation for θ10 kPa increased the explained variance to 86%, whereas expressing clay and silt contents on a volume basis instead of a mass basis increased the explained θ1.5 MPa variance to 83%. The avialable water storage capacity (AWC, here defined as moisture retained between 10 kPa and 1.5 MPa, θ10 kPa−1.5 MPa) showed significant correlations (at 1% level) with clay content, bulk density, and specific surface area, but not more than 48% of its variance could be explained. Parameters of the Van Genuchten equation showed significant (at 1% level) correlation with particle-size fractions, dithionite-extractable Fe and Al, organic-C content, and cation exchange capacity. Application of the estimated Van Genuchten equation to Set 1 resulted in a variance reduction of 79% for θ10 kPa, 75% for θ1.5 Mpa, and 45% for θ10 MPa. The results for Set 2 suggest that predicting AWC is most accurate when equations are used that are based on direct correlation between θ10 kPa−1.5 MPa and easily measured soil properties. Predictions based on th Van Genuchten equation, or on subtracting an equation for θ1.5 MPa from another one for θ10 kPa were inferior: they had even larger mean errors than predicting θ10 kPa−1.5 MPa as just the average value of Set 1.