Path analysis was used to test theoretical models of relations among variables known to predict differences in children's memory--strategies, capacity, and metamemory. Children in kindergarten to fourth grade (chronological ages 5 to 11) performed different memory tasks. Several strategies (i.e., sorting, clustering, rehearsal, and self-testing) were coded from videotapes and their average number of different strategies served as the strategy variable. Memory span and task-specific metamemory tasks served as the capacity and metamemory variables, respectively. The Wurzburg Model, derived from laboratories in Germany, had a significant chi-square statistic. The more successful Utilization Deficiency Model had a nonsignificant chi-square statistic and accounted for 48% of the variance in recall. Additional support for the Utilization Deficiency Model was evident from the significant path coefficients, which suggested that metamemory drives the production of strategies and that both metamemory and capacity constrain or enhance the effectiveness of those strategies. (Contains 6 figures and 4 tables.)