Multiple discrimination problems were presented concurrently to retarded children for 1, 2, 4, or 8 massed trials a day for a total of 16 trials per problem. Performance rose sharply over massed trials, then dropped during each 24-hr, retention interval. Long-term memory growth was related in a nonmonotonic manner to degree of massing. Two trials per day were most effective. Both the 1-trial and the 8-trial conditions produced significantly worse performance than the 2-trial and 4-trial conditions on long-term memory tests after equal numbers of training trials. A modified version of a Markov model, the generalall-or-none forgetting theory, predicted the results quite closely and was also shown to be capable of accounting for previously reported, but unexplained, nonmonotonic effects of spacing.