Viruses can be engineered to replicate in cancer cells selectively. This can be achieved using regulatory elements with tumor-selectivity, or using deletions of viral genes that are essential in normal cells but dispensable in tumor cells. The latter approach has been used to make viruses that depend on loss of RB or loss of p53. However, redundancy of function of viral proteins and cellular pathways has complicated this approach and required further viral engineering or a better understanding of cellular pathways to increase effectiveness. In spite of these complicating factors, ONYX-015, a prototype of this new class of therapeutic agents, has undergone extensive testing in the clinic, and has proven safe with promising signs of efficacy. Here, I summarize issues relating to selectivity of this agent and its clinical potential. Based on clinical data and new basic discoveries, several approaches are suggested to improve the efficacy of this agent in the clinic.