In support of the idea that certain RNA molecules might be able to catalyze RNA replication, a ribozyme was previously generated that synthesizes short segments of RNA in a reaction modeled after that of proteinaceous RNA polymerases. Here, we describe substrate recognition by this polymerase ribozyme. Altering base or sugar moieties of the nucleoside triphosphate only moderately affects its utilization, provided that the alterations do not disrupt Watson-Crick pairing to the template. Correctly paired nucleotides have both a lower K(m) and a higher k(cat), suggesting that differential binding and orientation each play roles in discriminating matched from mismatched nucleotides. Binding of the pyrophosphate leaving group appears weak, as evidenced by a very inefficient pyrophosphate-exchange reaction, the reverse of the primer-extension reaction. Indeed, substitutions at the gamma-phosphate can be tolerated, although poorly. Thio substitutions of oxygen atoms at the reactive phosphate exert effects similar to those seen with cellular polymerases, leaving open the possibility of an active site analogous to those of protein enzymes. The polymerase ribozyme, derived from an efficient RNA ligase ribozyme, can achieve the very fast k(cat) of the parent ribozyme when the substrate of the polymerase (GTP) is replaced by an extended substrate (pppGGA), in which the GA dinucleotide extension corresponds to the second and third nucleotides of the ligase. This suggests that the GA dinucleotide, which had been deleted when converting the ligase into a polymerase, plays an important role in orienting the 5'-terminal nucleoside. Polymerase constructs that restore this missing orientation function should achieve much more efficient and perhaps more accurate RNA polymerization.