Computer dialogs communicate important security messages, but their excessive use has produced habituation: a strong tendency by computer users to ignore security dialogs. Unlike physical warnings, whose design and use is regulated by law and based on years of research, computer security dialogs are often designed in an arbitrary manner. We need scientific solutions to produce dialogs that users will heed and understand. Currently, we lack an understanding of the factors that drive users’ attention to security dialogs, and how to counteract habituation. Studying computer security behavior is difficult because a) users are more likely to expose themselves to risk in a lab experiment than in daily life, b) the size of observed effects is usually very small, which makes it necessary to collect many observations, and c) it is complex to balance research interests and the ethical duty not to harm. My thesis makes two contributions: a novel methodology to study behavioral responses to security dialogs in a realistic, ethical way with high levels of ecological validity, and a novel technique to increase and retain attention to security dialogs, even in the presence of habituation.