Centriole duplication is a key step in the cell cycle whose mechanism is completely unknown. Why new centrioles always form next to preexisting ones is a fundamental question. The simplest model is that preexisting centrioles nucleate the assembly of new centrioles, and that although centrioles can in some cases form de novo without this nucleation, the de novo assembly mechanism should be too slow to compete with normal duplication in order to maintain fidelity of centriole duplication.We have measured the rate of de novo centriole assembly in vegetatively dividing cells that normally always contain centrioles. By using mutants of Chlamydomonas that are defective in centriole segregation, we obtained viable centrioleless cells that continue to divide, and find that within a single generation, 50% of these cells reacquire new centrioles by de novo assembly. This suggests that the rate of de novo assembly is approximately half the rate of templated duplication. A mutation in the VFL3 gene causes a complete loss of the templated assembly pathway without eliminating de novo assembly. A mutation in the centrin gene also reduced the rate of templated assembly.These results suggest that there are two pathways for centriole assembly, namely a templated pathway that requires preexisting centrioles to nucleate new centriole assembly, and a de novo assembly pathway that is normally turned off when centrioles are present.