The relationship between confidence and accuracy has long been an important and controversial topic within the field of human memory. In a recent review article, Schwartz (2024). "Inferential theories of retrospective confidence." Metacognition & Learning.) competently summarized some of the key empirical findings on this issue and clearly articulated two different extant theoretical approaches to understanding this relationship. The "direct access" view states that one's confidence in a memory is tied directly to the strength of the encoded memory trace, predicting a strong and near ubiquitous positive relationship between confidence and accuracy. In contrast, the "inferential view" holds that confidence is inferred from the heuristic use of available cues, and that any positive relationship between confidence and accuracy stems from the use of cues that correlate positively with accuracy. Here, I propose an alternative view that blends aspects of both accounts. Termed the "indirect access" account, I argue that memory signals and their experiential correlates form the basis of confidence judgments. This approach anticipates reported dissociations between confidence and accuracy, and accommodates a broad range of empirical findings. By this view, rare instances of weak confidence-accuracy relationships stem from strong misleading memory signals that experientially mimic strong accurate memory signals. Because strong memory signals are largely accurate, this view predicts a pervasive and robust positive relationship between confidence and accuracy.