Cancer 'genetic dependencies' - genes whose products are essential for cancer cell fitness - are promising targets for therapeutic development. However, recent evidence has cast doubt on the validity of several putative dependencies that are currently being targeted in cancer clinical trials, underscoring the challenges inherent in correctly identifying cancer-essential genes. Here we review several common techniques and platforms for discovering and characterizing cancer dependencies. We discuss the strengths and drawbacks of different gene-perturbation approaches, and we highlight the use of poorly validated genetic and pharmacological agents as a common cause of target misidentification. A careful consideration of the limitations of current technologies and cancer models will improve our ability to correctly uncover cancer genetic dependencies and will facilitate the development of improved therapeutic agents.