Measuring and combatting bias has been a central focus of policy and research in domains including employment, courts, news, medicine, college admissions, and AI. Yet notions of a what it means for a decision to be unbiased remain contested and often contradictory. I present a unified framework to define notions of bias and lack of bias across these and other domains, consider when “unbiased” is (or often isn’t) a helpful concept, relate these definitions to empirical evidence, and draw lessons for those seeking to combat harmful biases in society.