The article gives a part of an experiment on determining the most common errors in the digitization of objects of different topological types, namely considering errors in the setting of points. The experiment was performed in 2015–2017 on the basis of the Kursk State University, the number of participants was 63 people. Each of them digitized 30 points in four scales of the display (that is, 120 points from a subject). Then the directions and absolute values of the errors were determined. According to these initial data it was established that the widespread model of errors in the setting of points, defined by two-dimensional normal distribution, in many cases does not correspond to reality (was not observed with more than half of the subjects). The hypothesis of the equability of error directions was proved even to a lesser degree, since the majority had its own characteristic direction of error (most often this was displacement of the position of the point from the true to the southeast, least of all to the northwest). 45 out of 63 subjects demonstrated an uneven distribution of directions of setting points around the true position. The absolute value of the allowed errors in meters is proportional to the scale of the display, but is unchanged in pixels of the screen. The median error is approximately 1 pix, the 99th quantile of the distribution of error values corresponds to 4,1 pix. The results of the experiment make it possible to choose in advance the optimum scale of the digitization of the raster (not too large and not too shallow) in order to achieve the required quality of vectorization.