After two decades of research in parallel tools for performance tuning, why are users still dissatisfied? This paper outlines the human factors issues that determine how performance tools are perceived by users. The information provides insight into why current performance visualizations are not as well received as they should be—and what must be done in order to develop tools that are more closely aligned to user needs and preferences. Specific mechanisms are suggested for improving three aspects of performance visualizations: how the user explores the performance space, how the user compares different aspects of program behavior, and how the user navigates through complex source code.