Abstract: This paper propose a new sampling method, uniform design sampling （UDS）, which is a development of uniform design （UD） proposed by Wang anf Fanf （1981） and an improvement for Latin hypercube sampling （LHS） proposed by Mckay, conover and Beckman （1979）. We study the properties of this sampling and apply it to computations of numerical integrals, and compare it with Monte-Carlo method, UD （including URD）, and LHS （including OALHS）. The results show that UDS is more powerful than the other methods. Also, we give a discussion of how to use the UDS to general distribution cases.