Recombinant adeno-associated virus (rAAV) vectors are promising human gene transfer vectors, because they mediate long-term gene expression in vivo. The vector DNA form responsible for sustained gene expression has not been clearly defined, but it has been presumed that the vector integrates to some degree and persists in this manner. Using two independent methods, we were unable to identify rAAV integrants in mouse muscle. In the first approach, we were unable to recover host cell-vector DNA junctions from a lambda phage library generated using transduced mouse muscle DNA that contained a high vector copy number. Following this result, we devised a PCR assay based on the principle that integrated rAAV vector sequences could be amplified using primers specific for mouse interspersed repetitive sequences (B1 elements). Using this assay, we analyzed transduced mouse muscle DNA isolated from 6 to 57 weeks after injection and did not detect amplification above background levels. Based on the demonstrated sensitivity of the assay, these results suggested that >99.5% of vector DNA was not integrated. Additional analyses using a novel DNA exonuclease showed that the majority of the rAAV vector DNA in muscle persisted over time as transcriptionally active monomeric and concatameric episomes.