Theories of confidence judgments have embraced the role random error plays in influencing responses. An important next step is to identify the source(s) of these random effects. To do so, we used the stochastic judgment model (SJM) to distinguish the contribution of encoding and retrieval processes. In particular, we investigated whether dividing attention during encoding and/or retrieval changed the consistency of judgments during a statement verification task. Analysis of both choices and confidence ratings revealed: (1) divided attention at encoding increased the between-item variability and strength of the stored information thereby decreasing participants' ability to discriminate between true and false statements at test; and (2) divided attention at retrieval had no effect on item-level trial-by-trial variability, discrimination ability, or criterion location. We discuss the implications of these results in terms of theories of memory and of judgment. (Contains 6 figures and 5 tables.)