We elucidated molecular mechanism of RNAP II pausing on AP site and transcription bypass. We found two consecutive RNAP II stalled states that correspond to slow down insertion and extension steps. We biochemically characterized and solved the structures of the AP-arrested polymerase in the absence and presence of bound NTP, representing the states when RNAP II first encounters the AP site. In the first stalled state, we found that the RNA:DNA hybrid is in a canonical post-translocation state register, but the AP site is located above the Bridge Helix as intermediate state. Because AP lacks nucleobase group and therefore AP site cannot support effective template-dependent NTP insertion. As a result, the polymerization rates and selectivity of NTP incorporation are significantly decreased. We observed that AMP (followed by GMP) are preferentially incorporated ('A-rule'). We further biochemically characterized and solved RNAP II structure at the AP lesion with AMPCPP bound. We found that AMPCPP binds to the addition site opposite the AP lesion, suggesting the base stacking is likely the driving force for ATP binding as a foundation for the 'A-rule'. To understand the slow extension step, we solved the structure of RNAP II after AMP is incorporated opposite the AP lesion. In this structure, RNAP II occupied post-translocation register in which we observed a significant flexibility for both AMP at 3'-RNA terminus and 5'-template nucleotide adjacent to the AP lesion, because of a lack of stacking interactions and base pair restraint. As a result, the position of incoming NTP is also compromised, as shown in EC-IV structure, we found that the cognate UTP fails to reach A-site, and largely remains at E-site. Moreover, the instability of DNA/RNA scaffold with AMP against the AP site promotes Pol II backtracking. These structures provide structural explanation to why extension step is also very slow and why the extension step does not obey the A-Rule. We speculate that, occasionally, when the template base (dA) moves into canonical i+1 template position and UTP can rotate into A-site and form Watson-Crick base pair and proceed template-dependent addition, whereas ATP would be rejected due to mismatched base pair. Once UTP is incorporated into RNA and Pol II moves forward (n+2), the AP lesion would be sandwiched by non-damaged Watson-Crick base pair. The impact of AP site for downstream nucleotide binding and incorporation would be minimal, as revealed by our biochemical data in the absence and presence of TFIIS. The downstream Pol II elongation is similar to the transcription from non-damaged template.