A sample drawn in several phases refers to the selection of a large-sized preliminary sample on which some quick and inexpensive method of measurement is applied,-followed by the drawing of a subsample on which a somewhat more elaborate method of measurement is used. Then successively nested subsamples are drawn and more expensive, elaborate and accurate methods of measurement are applied. This paper furnishes expressions for the estimator and its variance for the three-phase design for both quantitative and binary data. An example from a remote sensing application is presented to illustrate estimation of the population mean of a numerical characteristic. Monte Carlo data are used to illustrate the estimation of the population proportion of a binary variable of interest. Formulas for optimum sampling rates are shown to be of the same form for the two types of variables.