Rats were run in a Y maze; responses to one side were followed by food after a constant d sec. of delay, while responses to the other side were rewarded after delays of either 0 or 2d sec., presented in a random sequence. A preference for the variable-delay side developed, and developed more rapidly the larger the value of d, confirming predictions which were derived from the assumption of a negatively accelerated gradient of reinforcement. (PsycINFO Database Record (c) 2006 APA, all rights reserved)