RNA interference has become widely used as an experimental tool to study gene function. In addition, small interfering RNA (siRNA) may have great potential for the treatment of diseases. Recently, it was shown that siRNA can be used to mediate gene silencing in mouse models. Locally administered siRNAs entered the first clinical trials, but strategies for successful systemic delivery of siRNA are still under development. Challenges still exist about the stability, delivery, and therapeutic efficacy of siRNA. In the present study, we compare the efficacy of two methods of systemic siRNA delivery and the effects of siRNA modifications using locked nucleic acids (LNA) in a xenograft cancer model. Low volume tail vein bolus injections and continuous s.c. delivery using osmotic minipumps yielded similar uptake levels of unmodified siRNA by tumor xenografts. Both routes of administration mediated sequence-specific inhibition of two unrelated targets inside tumor xenografts. Previous studies have shown that LNA can be incorporated into the sense strand of siRNA while the efficacy is retained. Modification of siRNA targeting green fluorescent protein with LNA results in a significant increase in serum stability and thus may be beneficial for clinical applications. We show that minimal 3' end LNA modifications of siRNA are effective in stabilization of siRNA. Multiple LNA modifications in the accompanying strand further increase the stability but negate the efficacy in vitro and in vivo. In vivo, LNA-modified siRNA reduced off-target gene regulation compared with nonmodified siRNA. End-modified siRNA targeting green fluorescent protein provides a good trade-off between stability and efficacy in vivo using the two methods of systemic delivery in the nude mouse model. Therefore, LNA-modified siRNA should be preferred over unmodified siRNA.