Confidence in memory decisions usually mirrors the accuracy of those decisions, but there are exceptions. One exception is believed to occur when external cues - such as information from other people - are introduced during source memory tasks. While these external cues do influence accuracy depending on their validity, confidence judgements (CJs) remain unchanged. In a series of experiments, we set out to verify the robustness of the confidence-accuracy dissociation pattern in source memory tasks. First, we replicated the dissociation with a source task in which participants studied individually presented words (items) together with orienting questions (sources). We then manipulated source strength by varying the number of presentations of study items (three versus one) and assigning separate sources to strong versus weak items. Finally, we eliminated the confound by which item and source strength were correlated by keeping the number of item presentations constant while manipulating the number of presentations of each item with an orienting question (three out of three presentations versus one out of three). Together, we have shown that although the confidence-accuracy dissociation generally holds in source memory tasks, it is not universally found, and we present boundary conditions for its appearance.