Prediction of soil attributes (properties and classes) in digital soil mapping is based on the correlation between primary soil attributes and secondary environmental attributes. These secondary attributes can be obtained relatively cheaply over large areas. In the presence of these environmental covariates, a strategic sampling design can ensure the coverage of the full range of environmental variables. This could enhance the full representation of the expected soil properties or soil classes. This paper will present Latin hypercube sampling (LHS) as a sampling strategy on existing data layers. LHS is a stratified-random procedure that provides an efficient way of sampling variables from their multivariate distributions. LHS involves sampling N_s values from the prescribed distribution of each of the variables. The cumulative distribution for each variable is divided into N_s equiprobable intervals, and a value is selected randomly from each interval. The N_s values obtained for each variable are then paired with the other variables. This method ensures a full coverage of the range of each variable by maximally stratifying the marginal distribution. The second part of this paper deals with filling the spatial gaps in a previously surveyed area. This technique imposes a regular grid on previously surveyed area and attempts to find additional points in the grid that can provide a better coverage on the area. These methods will be illustrated with examples from digital soil mapping of part of the Hunter Valley of New South Wales.