Salmonella has a natural ability to target a wide range of tumors in animal models. However, strains used for cancer therapy have generally been selected only for their avirulence rather than their tumor-targeting ability. To select Salmonella strains that are avirulent and yet efficient in tumor targeting, a necessary criterion for clinical applications, we measured the relative fitness of 41,000 Salmonella transposon insertion mutants growing in mouse models of human prostate and breast cancer. Two classes of potentially safe mutants were identified. Class 1 mutants showed reduced fitness in normal tissues and unchanged fitness in tumors (e.g., mutants in htrA, SPI-2, and STM3120). Class 2 mutants showed reduced fitness in tumors and normal tissues (e.g., mutants in aroA and aroD). In a competitive fitness assay in human PC-3 tumors growing in mice, class 1 mutant STM3120 had a fitness advantage over class 2 mutants aroA and aroD, validating the findings of the initial screening of a large pool of transposon mutants and indicating a potential advantage of class 1 mutants for delivery of cancer therapeutics. In addition, an STM3120 mutant successfully targeted tumors after intragastric delivery, opening up the oral route as an option for therapy administration.