BACKGROUND: When designing therapeutic short-interfering RNAs (siRNAs), off-target effects (OTEs) are usually predicted by computational quantification of messenger RNAs (mRNAs) that contain matches to the siRNA seed sequence in their 3' UTRs. It is assumed that the higher the number of predicted transcriptional OTEs, the greater the size of the actual OTE signature and the more detrimental the phenotypic consequences in target-negative cells. METHODS: We tested this general assumption by investigating the OTEs of potential therapeutic siRNAs targeting the human papillomavirus (HPV) type-16 E7 oncogene. We studied HPV-negative squamous epithelial cells, from normal cervix (NCx) and skin (HaCaT), which would be vulnerable to 'bystander' OTEs following transfection in vivo. RESULTS: We observed no correlation between the number of computationally predicted OTEs and the actual number of seed-dependent OTEs (P=0.76). On average only 20.5% of actual transcriptional OTEs were seed-dependent (i.e., predicted). The unpredicted OTEs included stimulation of innate immune pathways, as well as indirect (downstream) effects of other OTEs, which affected important cancer-associated pathways. Although most significant OTEs observed were seen in both NCx and HaCaT cells, only 0-5.9% of differentially expressed genes overlapped between the two cell types. CONCLUSION: These data do not support the assumption that actual OTEs correlate well with predicted OTEs.