Accurate calculation of mutation rates for viruses and viroids is necessary for evolutionary studies and to evaluate adaptation potential. However, estimation of in vivo mutation rates is complicated by selection, which leads to loss or proliferation of certain mutations. To minimize this concern, lethal mutations, including nonsense and non-synonymous mutations, have been used to determine mutation rates for several viruses and viroids, including Potato spindle tuber viroid (PSTVd). However, this approach has limitations, including focus on a relatively small number of genome sites and the possibility that mutations may not actually be lethal or may be maintained by wild type individuals. To avoid selection bias altogether, we sequenced minus-strand PSTVd dimers from concatemeric replication intermediates. The underlying rationale is that mutations found in only one of the monomers were likely generated de novo during RNA polymerase II (Pol II) transcription of the circular plus-strand RNA genome. This approach yielded an apparent Pol II error rate of ~1/1837 nucleotides per transcription cycle, and an estimated mutation rate of ~1/919 nucleotides for a single replication cycle. Remarkably, de novo mutations were nearly absent from the most conserved, replication-critical regions of the PSTVd genome, suggesting that sequence conservation is a consequence of both essential function and template optimization for greater Pol II fidelity. Such biased fidelity may constitute a novel strategy to ensure population success while allowing abundant sampling of sequence space in other genome regions. Comparison with variants in progeny populations derived from a cloned, wild type PSTVd master sequence revealed that most de novo mutations were lost through selection.