Based on the work of Wang and Fang (1990a, b), this paper extends the def-inition of discrepancy of a set of points with respect to a distribution. To construct multivariate uniform designs, two transformations, discrepancy-preserving transfor-mation and density-preserving transformation, and a method of separating variables are introduced. Also, for some special domains useful in statistics, some related transformations are given. In particular, as an application of the method of sepa-rating variables, we give an approach of constructing a uniform design in the Stiefel manifold and outline its applications in projection pursuit methods.