The quality of statistical inference is dependent not only on e.g. estimator construction but on the structure of a population and a sampling scheme, too. For example, let the estimation of total wheat production in a population of farms be considered. The population of farms is divided into clusters corresponding to villages. This estimation can be based on the ordinary simple sample or on the cluster sample. Population units can be selected to the sample by means of several sampling schemes. The units (i.e. farms) can be selected to the ordinary sample, or clusters of the units (i.e. villages) can be drawn to the cluster sample. The accuracy of the estimation depends on the sampling scheme and on the intra-class spread of a variable under study.