Metamemory effects under retroactive interference (RI) were tested using a modified RI paradigm in order to determine whether relatedness of word pairs impacts metamemory predictions and to investigate the bases of these predictions. Conflicting findings from prior research suggested that the relatedness of materials used to test RI could produce disparate effects; some studies showed association between memory and metamemory but other studies showed dissociation between the two. The experiments consisted of four phases: original and interpolated study, prediction, and test; participants were tested on the words encountered at original study. By comparing predictions to recall, we determined that memory and metamemory were dissociated under RI regardless of cue-target relatedness and regardless of the relatedness of targets from original to interpolated study. Additionally, the findings support the use of the accessibility heuristic to make metamemory predictions under RI. The results support a separate systems viewpoint.