Unconscious transference (UT) is a potential explanation for an eyewitness identifying a familiar but innocent suspect from a lineup. We conducted a large experiment (N = 31,612) to test three UT theories (automatic processing/familiarity, poor source retrieval, and memory blending) by manipulating memory strength for the perpetrator versus the bystander, as well as their physical similarity. Participants were randomly assigned to view a fair simultaneous lineup containing the perpetrator, the bystander, an innocent suspect never seen before, or both the perpetrator and bystander. We found several forms of evidence of UT, but it was much weaker when: (a) the bystander was dissimilar from the perpetrator, and (b) the perpetrator was viewed for longer than the bystander. Both automatic processing/familiarity and poor source retrieval received the most support, with memory blending a distant third. Last, UT harmed the confidence-accuracy relationship. We recommend additional eyewitness research involving familiar innocent suspects.