摘要：It is well understood tha. for conventiona. survey designs the set of unordered distinct units in a.sa.ple is a.minima.ly sufficient sta.istic. This mea.s tha. for inferentia. sta.istic of the sa.ple, the va.ue of the sa.pled units ra.her tha. the sa.ple design is importa.t. Sa.pling ra.e popula.ions presents distinct cha.lenges. Exa.ples of ra.e popula.ions a.e in biology with ra.e a.d enda.gered a.ima.s where there a.e only a.few rema.ning individua.s, or in socia. science, with the low incidence of people from a. unusua.ly high （or low） income group. Sa.pling ra.e popula.ions tends to result in the ca.e tha. ma.y of the sa.ple units do not conta.n informa.ion on the cha.a.teristic of interest （e.g., the ra.e a.ima., or people from the unusua. income group）. For finite ra.e popula.ions the set of unordered distinct ra.e-units in a.sa.ple is a.minima.ly sufficient sta.istic. In a. exa.ple ca.e study of a.ra.e buttercup, the properties of the minima. sufficient estima.or a.e explored. We compa.e the efficiency of the estima.or for the popula.ion tota. ba.ed on the minima.ly sufficient sta.istic, with the sta.da.d estima.or for a.ra.ge of sa.ple sizes. The va.ia.ce of the minima.ly sufficient estima.or wa. a.wa.s sma.ler tha. the va.ia.ce of the sufficient estima.or. For ra.e popula.ions where non-ra.e units ca. be distinguished from ra.e units beca.se they ha.e the sa.e fixed va.ue, the minima. sufficient sta.istic is the ra.e units, if a.y, in the sa.ple.