Experiment I was designed to find a functional relationship between the number of reinforced trials in terms of learning criteria and the easiness of shifting to a new discrimination learning in young children. Group 5 was required to learn a position discrimination until 5 successive correct responses, and Group 10 until 10 successive rights and Groups 15, 20 and 30were likewise required to learn the problem until corresponding numbers of successes. As soon as they met the criteria, the problem was shifted to color(white vs.red)discrimination with the same criteria of 10 successive correct responses for all groups. Group C10 was added, which was the reverse of Group 10:they were given position discrimination after color problem. The results indicated that Group 10 required the greatest number of trials for learning the second color discrimination, and along which a bi-directional gradient was demonstrated in terms of trials, although the overall group differences were not statistically significant(Table 1, Fig.1). Number of perseverative errors from position learning was found to be positively related to difficulty of shifting to color discrimination. In Experiment II replicated Experimet I with slight modifications, in that the subjects used were a little older and although Group 5 was exactly the same as Group 5 in the previous experiment, Groups 10, 20 and 40 were given additional 5, 15 and 35 trials over the criteria used in Group 5. Again a similar finding was obtained(Table 3, Fig.1). These results failed to show an agreement with previous studies where the positive effects of the overlearning upon the easiness of shifts of post learning were reported. The present writers suggests two factors in explaining the results. The first is the response habit connected to a particular stimulus situation as is assumed by the orthodox S-R theorists. This habit evolves rapidly with trials and its strength is negatively related to the easiness of discrimination shiftings. The difference between Groups 5 and 10 is accounted for by this factor. The second facor of discrimination set, on the other hand, develops only after the habit is fairly established. As this factor is trans-situational, it makes shifts of learning easier. In addition, if Krechevsky's concept of "hypothesis' is true during the pre-solution perid, then this factor togeher with the two factrs previously mentioned can explain most part of discrimination learning(Fig.2).