This paper introduces what is called the H-technique for building cumulative scales with increased precision. The method simply consists of determining a given cutting point in a Guttman or Lazarsfeld latent distance scale not by means of a single response but rather by means of several responses, which are formed into a new “contrived item.” The objective is to maximize the information available from the basic data, and hence to strengthen confidence in the scalability of the area under consideration and the generality of the dimension which the scale is definning, and to improve the ranking of individuals through reduction of scale error. A theorectical example shows the advantages in a special case in which the structure of the population is completely specified. An empirical example from actual data illustrates the procedures used in the computing.