Sidmanian avoidance training in rats was performed with a revised apparatus and some considerablly effective learnings were observed through three experiments. A shock-transmitter mechanism made with ball bearings brought succesively scrambled AC 20 V shocks to a grid conforming the surface of a rotating cage. A same avoidance training schedule was used in the three experiments. This is shock-shock interval was 5s, response-shock interval was 20s, and a shock duration was 0.5s. In Experiment 1, the main variable was the magnitude of rotation to be counted as a response, with two variables of a half and a quarter of one round, respectively. Although under both conditions rats could easily acpuired avoidance responses, the criterion of a quarter rotation was more effective than that of a half. In Experiment 2, the maintenance of learning over night was tested. Rats showed some superior maitenance of avoidance response without any over-night decrement for consecutive 8 days. Rats ran a mean of 504 m/h continuousely, and number of shocks recieved reduced to 0-3 times/h at the last day. In Experiment 3, the continuation of avoidance responses through a long session, using rats which had already learned the avoidance responses. Avoidance responses were continued until the end of the session over 6 hours, and animals ran a mean of 1544m with a mean of 42 shocks during 6 hours.