Hepatitis B virus (HBV) is a promising target for therapies based on RNA interference (RNAi) since it replicates via RNA transcripts that are vulnerable to RNAi silencing. Clinical translation of RNAi technology, however, requires improvements in potency, specificity and safety. To this end, we systematically compared different strategies to express anti-HBV short hairpin RNA (shRNA) in a pre-clinical immunocompetent hepatitis B mouse model. Using recombinant Adeno-associated virus (AAV) 8 vectors for delivery, we either (i) embedded the shRNA in an artificial mi(cro)RNA under a liver-specific promoter; (ii) co-expressed Argonaute-2, a rate-limiting cellular factor whose saturation with excess RNAi triggers can be toxic; or (iii) co-delivered a decoy ("TuD") directed against the shRNA sense strand to curb off-target gene regulation. Remarkably, all three strategies minimised adverse side effects as compared to a conventional shRNA vector that caused weight loss, liver damage and dysregulation of > 100 hepatic genes. Importantly, the novel AAV8 vector co-expressing anti-HBV shRNA and TuD outperformed all other strategies regarding efficiency and persistence of HBV knock-down, thus showing substantial promise for clinical translation.