This is the final published version. It first appeared at http://journals.plos.org/plosgenetics/article?id=10.1371/journal.pgen.1005078. ; In the nematode Caenorhabditis elegans, different small RNA-dependent gene silencing mechanisms act in the germline to initiate transgenerational gene silencing. Piwi-interacting RNAs (piRNAs) can initiate transposon and gene silencing by acting upstream of endogenous short interfering RNAs (siRNAs), which engage a nuclear RNA interference (RNAi) pathway to trigger transcriptional gene silencing. Once gene silencing has been established, it can be stably maintained over multiple generations without the requirement of the initial trigger and is also referred to as RNAe or paramutation. This heritable silencing depends on the integrity of the nuclear RNAi pathway. However, the exact mechanism by which silencing is maintained across generations is not understood. Here we demonstrate that silencing of piRNA targets involves the production of two distinct classes of small RNAs with different genetic requirements. The first class, secondary siRNAs, are localized close to the direct target site for piRNAs. Nuclear import of the secondary siRNAs by the Argonaute HRDE-1 leads to the production of a distinct class of small RNAs that map throughout the transcript, which we term tertiary siRNAs. Both classes of small RNAs are necessary for full repression of the target gene and can be maintained independently of the initial piRNA trigger. Consistently, we observed a form of paramutation associated with tertiary siRNAs. Once paramutated, a tertiary siRNA generating allele confers dominant silencing in the progeny regardless of its own transmission, suggesting germline-transmitted siRNAs are sufficient for multigenerational silencing. This work uncovers a multi-step siRNA amplification pathway that promotes germline integrity via epigenetic silencing of endogenous and invading genetic elements. In addition, the same pathway can be engaged in environmentally induced heritable gene silencing and could therefore promote the inheritance of acquired traits. ; This study was supported by funding from: Cancer Research UK (http://www.cancerresearchuk. org), grant RG57329 to EAM; European Research Council (erc.europa.eu/) Framework Programme 7, grant RG58558 to EAM; Gonville and Caius College fellowship to PS; Career Development Award from the Medical Research Council (http://www.mrc.ac.uk/) to PS. Some strains were provided by the CGC, which is funded by NIH Office of Research Infrastructure Programs (P40 OD010440).