One of the most important statistical developments in recent years has been the great increase in the amount of attention paid to measuring the amount of error present in data collected in censuses and sample surveys. One reason for this increased emphasis is undoubtedly the great expansion in the number and types of sample surveys, which are usually designed with reference to the degree of precision required. In planning such surveys, the problem is to measure the total error of a statistical figure including both sampling and non-sampling errors. Important pioneering work in this field has been done by Mahalanobis and, more recently, by workers in France, Great Britain and the United States. As a result, more and more users of statistics have come to realize that certain sources of error exist, whether the survey is based upon a small sample or covers the entire universe. We have learned indeed that in certain cases the most accurate measure of a particular item may be obtained through a relatively small sample survey where it is economically feasible to put great emphasis upon reduction of the non­sampling error through careful training and follow-up. The decrease in non­sampling error thus achieved will often more than counterbalance the increase in sampling error. This paper, then, has the objective of calling attention to the techniques of measuring non-sampling error employed by the Bureau of the Census and to some of the results obtained by the use of these techniques. We shall deal with two general lines of approach: 1) the study of statistical totals and 2) the study of response errors.