25/27 Base duplex RNAs that are substrates for Dicer have been demonstrated to enhance RNA interference (RNAi) potency and efficacy. Since the target sites are not always equally susceptible to suppression by small interfering RNA (siRNA), not all 27-mer duplexes that are processed into the corresponding conventional siRNAs show increased potency. Thus random designing of Dicer-substrate siRNAs (DsiRNAs) may generate siRNAs with poor RNAi due to unpredictable Dicer processing. Previous studies have demonstrated that the 3′-overhang affects dicing cleavage site and the orientation of Dicer entry. Moreover, an asymmetric 27-mer duplex having a 3′ two-nucleotide overhang and 3′-DNA residues on the blunt end has been rationally designed to obtain greater efficacy. This asymmetric structure directs dicing to predictably yield a single primary cleavage product. In the present study, we analyzed the in vitroand intracellular dicing patterns of chemically synthesized duplex RNAs with different 3′-overhangs. Consistent with previous studies, we observed that Dicer preferentially processes these RNAs at a site 21–22 nucleotide (nt) from the two-base 3′-overhangs. We also observed that the direction and ability of human Dicer to generate siRNAs can be partially or completely blocked by DNA residues at the 3′-termimi. To examine the effects of various 3′-end modifications on Dicer processing in cells, we employed Illumina Deep sequencing analyses to unravel the fates of the asymmetric 27-mer duplexes. To validate the strand selection process and knockdown capabilities we also conducted dual-luciferase psiCHECK reporter assays to monitor the RNAi potencies of both the “sense” (S) and “antisense” (AS) strands derived from these DsiRNAs. Consistent with our in vitro Dicer assays, the asymmetric duplexes were predictably processed into desired primary cleavage products of 21–22-mers in cells. We also observed the trimming of the 3′ end, especially when DNA residues were incorporated into the overhangs and this trimming ultimately influenced the Dicer-cleavage site and RNAi potency. Moreover, the observation that the most efficacious strand was the most abundant revealed that the relative frequencies of each “S” or “AS” strand are highly correlated with the silencing activity and strand selectivity. Collectively, our data demonstrate that even though the only differences between a family of DsiRNAs was the 3′ two-nuclotide overhang, dicing polarity and strand selectivity are distinct depending upon the sequence and chemical nature of this overhang. Thus, it is possible to predictably control dicing polarity and strand selectivity via simply changing the 3′-end overhangs without altering the original duplex sequence. These optimal design features of 3′-overhangs might provide a facile approach for rationally designing highly potent 25/27-mer DsiRNAs.