The task considered here is the perceptual recognition of objects which are graphically represented in an incomplete way, because of the figural superposition of a grid that partly occludes their contours. Three kinds of actual training are defined, differing from each other in the graphical representation of the covering grids in the stimulus-figures; the fourth kind of training is a fictitious one, as it makes use of pictures with complete objects (i.e., without any covering grid). The experiment involved 76 children, mean age 5 years and 6 months; the whole sample was divided into four homogeneous groups, which were subjected to the four different kinds of training. Data from the experiment show that training is generally beneficial to the perceptual recognition task (higher performances are obtained following non-fictitious trainings), and point to a consistent relation between difficulty of the training procedure and improvement of performance at testing. Data also show that strategies learned during the training are not strictly specific to the stimuli used, and that a general "transfer of learning" took place in the processes examined in the experiment.