The functional deficit underlying amnesia has been argued to be in recollective processing. This idea is based on the DPSD model, wherein recognition comprises a mixture of recollection and familiarity signals, with familiarity conforming to an equal-variance signal-detection mechanism while recollection is binary. This model interprets the greater variance for targets than for lures revealed in sub-unit zROC slopes, to be a consequence of the mixture of two signals. Importantly, equal variance between targets and lures is found in amnesic, and is consequently interpreted by DPSD to reflect impairment to recollection alongside the sparing of familiarity. Here, we pointed to a logical fallacy in this interpretation. We then asked participants, in two experiments, to make remember-know (RK) and confidence judgments. Simulating equal variance in healthy participants, we either excluded from the analysis ‘remember’ responses, reflecting recollection, or the most accurate memories, reflecting strength. We found that only the exclusion of the strongest responses led to equal-variance distributions. In addition, we found that accuracy was associated with an interlaced ordering of RK response groups nested under confidence, a pattern hard to reconcile with classic recognition models (DPSD, UVSD). This pattern can, however, be accommodated by the Continuous Dual Process (CDP) model (Wixted and Mickes, 2010), wherein both familiarity and recollection are continuous signals. Amnesia may thus be characterized as the inability to form strong memories, recollection as well as familiarity.