To test the reliability of four tasks of the Michigan Restricted Association Norms--Superordinates, Parts, Following Words, and Free Associations--30 subjects were asked to respond to 33 stimulus items under each of the four instructions and to perform identical tasks 7-14 days later. Comparisons of the intra-individual reliability and inter-individual response distributions were made between tasks, frequency of stimuli, and sessions. Distributions differed between tasks but not between stimulus frequencies or sessions. Reliability was highest for Superordinates, slightly lower for Free Associations and Parts, and lowest for Following Words. The absolute magnitude of reliability was regarded as sufficiently high though subject to purpose-specific evaluation. Attempts also were made to compare the differences in reliability with differences in the response distribution of the norms. For Restricted but not for Free Associations the strength of the primary responses in the Michigan Norms correlates with the reliability coefficients. (KB)