Predicting accurate bond-length alternations (BLAs) in long conjugated molecular chains has been a major challenge for electronic-structure theory for many decades. While Hartree-Fock (HF) overestimates BLA significantly, second-order perturbation theory and commonly used density functional theory (DFT) approaches typically underestimate it. Here, we discuss how this failure is related to the many-electron self-interaction error (MSIE), which is inherent to both HF and DFT approaches. We use tuned long-range corrected hybrids to minimize the MSIE for a series of polyenes. The key result is that the minimization of the MSIE alone does not yield accurate BLAs. On the other hand, if the range-separation parameter is tuned to yield accurate BLAs, we obtain a significant MSIE that grows with chain length. Our findings demonstrate that reducing the MSIE is one but not the only important aspect necessary to obtain accurate BLAs from density functional theory.