Multi-dimensional scaling has been used to analyze the data on discrimination between configurative features of symbols (Russian letters, Arabic letters, some artificial signs) with the view to establish the minimal number of features needed to a man for discrimination. Interpretative analysis of the obtained structures of the subjective sign discrimination space reveals three configurative features determining the discrimination: the size of angles between lines composing the configuration, presence of local closed figures in the configuration, visual complexity of the configuration measured by the number of angle-producing lines. A hypothesis is forwarded that in man's visual system there exist gradual predetector neurons responsible for detection of basal features of configurations.