This report investigates children's analogical reasoning in a physics task, using an analogy generated by the children rather than by the experimenter. A total of 127 elementary school children took part in three related studies. Children learned to predict the behavior of a balance scale. Later, they were asked to solve a force interaction problem. Two versions of the balance scale training were devised: version A suggested an incorrect solution to the target problem (negative analogy), and version B suggested a correct solution to the target problem (positive analogy). In Study 1, 9- to 10-year-olds showed spontaneous transfer in both training conditions. In Study 2, 7-year-olds did not show any transfer in the positive analogy condition. Study 3 revealed that the lack of transfer in younger children was not due to a failure either to notice the analogy or to perform the mapping. Instead, 7-year-olds transferred only selected aspects of the correct solution.