RNA polymerases are molecular machines of great fidelity, which can recognize matched NTPs from unmatched NTPs and 2'-dNTPs. We investigated by a stochastic simulation algorithm the whole nucleotide addition cycle based on an event-driven model. This model allows us to examine possible molecular origins of the high fidelity of RNA polymerases. For unmatched NTP selectivity, the conclusions drawn from simulated elongation rates corroborate those derived from structural analysis. The presence of two conformations (Esite and pre-insertion site) for the incoming nucleotide before the polymerization reaction is sufficient to allow selectivity. Concerning sugar selectivity, our results indicate that selectivity is only achievable if slow chemical reactions occur for 2'-dNTP. These results can be used to understand recent experimental observations.