In this paper, we propose a novel coding scheme, which is referred to as twisted-pair superposition transmission (TPST) and can be constructed from any given basic code by “mixing together” a pair of basic codewords in a “twisted” manner. We present a successive cancellation list decoding algorithm for TPST, where a list of candidates for the first layer is generated serially and the most competitive one is identified by combining the second layer. Thresholds on empirical divergence function (EDF) are introduced for early termination to trade off performance with decoding complexity. Genie-aided bounds are derived, indicating that the performance of TPST codes can be improved by employing partial superposition. Numerical simulation results show that, by taking tail-biting convolutional codes (TBCCs) as basic codes, we can construct TPST-TBCCs with near-capacity performance in the short length regime. The construction is flexible in the sense that it can be easily adapted to a wide range of coding rates.