Abstract: The horses were required to differentiate a box containing a black cloth from two associated boxes with no cloth. The correct box contained food. It differed in actual position with respect to the other boxes from trial to trial. After learning had occurred, the position of the cloth signal was changed. The changed positions produced many errors, the number of errors depending upon whether the cloth was higher or lower than the opening of the food box. Retests showed original learning to be relatively stable and unaffected. Factors influencing accuracy of discrimination were: contacts with cloth, position of box with respect to entrance, age of the horse (the younger made fewer errors), and breed and type of horse. There is evidence of some retention after three years. (PsycINFO Database Record (c) 2012 APA, all rights reserved)