Semiconductor junctions are at the heart of many semiconductor devices from the simplest diodes to high performance transistors and photonic devices. Among devices based on a doped p-n or p-i-n junction, the polymer light-emitting electrochemical cell (PLEC) is an unique solid-state device possessing attractive attributes for low-cost applications, but also a junction structure that is still poorly understood. In a PLEC, the p-n junction is formed by applying a voltage bias between the contacts of the device. The applied voltage causes in situ electrochemical p-and n-doping of the semiconducting polymer and the formation of a dynamic light-emitting p-n junction. Once the junction is fixed by cooling or chemical manipulation, the "frozen-junction" PLEC exhibits an unipolar electroluminescence (EL) and photovoltaic response. Repeated thermal cycling, however, can cause the frozen-junction PLEC to experience drastically enhanced EL under forward bias and the emergence of reverse bias EL. In this study, we use a combination of transport measurements and direct imaging to elucidate the origin of the mysterious reverse bias EL. We develop a model that explains the reverse bias EL as caused by the tunnel injection of electrons and holes from band gap states into a de-doped "intrinsic" region between the p-and n-doped regions. The model explains the location, relative intensity and evolution of EL under both forward and reverse bias. The results strongly hint at a junction that is much narrower than previously resolved.