RNA interference (RNAi) in both plants and animals is mediated by small RNAs of approximately 21-23 nucleotides in length for regulation of target gene expression at multiple levels through partial sequence complementarities. Combined with widespread genome sequencing, experimental use of RNAi has the potential to interrogate systematically all genes in a given organism with respect to a particular function. However, owing to a tolerance for mismatches and gaps in base-pairing with targets, small RNAs could have up to hundreds of potential target sequences in a genome, and some small RNAs in mammalian systems have been shown to affect the levels of many messenger RNAs besides their intended targets. The use of long double-stranded RNAs (dsRNAs) in Drosophila, where Dicer-mediated processing produces small RNAs inside cells, has been thought to reduce the probability of such 'off-target effects' (OTEs). Here we show, however, that OTEs mediated by short homology stretches within long dsRNAs are prevalent in Drosophila. We have performed a genome-wide RNAi screen for novel components of Wingless (Wg) signal transduction in Drosophila S2R + cells, and found few, if any, legitimate candidates. Rather, many of the top candidates exert their effects on Wg response through OTEs on known pathway components or through promiscuous OTEs produced by tandem trinucleotide repeats present in many dsRNAs and genes. Genes containing such repeats are over-represented in candidate lists from published screens, suggesting that they represent a common class of false positives. Our results suggest simple measures to improve the reliability of genome-wide RNAi screens in Drosophila and other organisms.