Probabilistic retroactive interference (RI) refers to the interfering effects of intermixing presentations of an earlier studied response (A-B) with presentations of a competing response (A-D). As an example, for a 2/3 condition, a cue word was presented with its earlier studied response twice and its competing response once during the interference phase. Performance on direct and indirect tests of memory for earlier studied responses was combined to reveal dissociations between effects on recollection and accessibility bias. Manipulating probabilistic RI influenced accessibility bias but left recollection unchanged. Effects of probabilistic RI were compared with effects of traditional, nonprobabilistic RI. The authors contrast their dual-process model with traditional accounts of RI and discuss the importance of distinguishing between recollection and accessibility bias for understanding interference effects.