Reactive oxygen species (ROS) and reactive nitrogen species (RNS) are generated from aerobic metabolism, as a result of accidental electron leakage as well as regulated enzymatic processes. Because ROS/RNS can induce oxidative injury and act in redox signaling, enzymes metabolizing them will inherently promote either health or disease, depending on the physiological context. It is thus misleading to consider conventionally called antioxidant enzymes to be largely, if not exclusively, health protective. Because such a notion is nonetheless common, we herein attempt to rationalize why this simplistic view should be avoided. First we give an updated summary of physiological phenotypes triggered in mouse models of overexpression or knockout of major antioxidant enzymes. Subsequently, we focus on a series of striking cases that demonstrate "paradoxical" outcomes, i.e., increased fitness upon deletion of antioxidant enzymes or disease triggered by their overexpression. We elaborate mechanisms by which these phenotypes are mediated via chemical, biological, and metabolic interactions of the antioxidant enzymes with their substrates, downstream events, and cellular context. Furthermore, we propose that novel treatments of antioxidant enzyme-related human diseases may be enabled by deliberate targeting of dual roles of the pertaining enzymes. We also discuss the potential of "antioxidant" nutrients and phytochemicals, via regulating the expression or function of antioxidant enzymes, in preventing, treating, or aggravating chronic diseases. We conclude that "paradoxical" roles of antioxidant enzymes in physiology, health, and disease derive from sophisticated molecular mechanisms of redox biology and metabolic homeostasis. Simply viewing antioxidant enzymes as always being beneficial is not only conceptually misleading but also clinically hazardous if such notions underpin medical treatment protocols based on modulation of redox pathways.