At the world's some 440 nuclear power plants, experts continually monitor their wide safety margins, and at signs of trouble seek out the sources and recommend changes. Too often for their comfort, and for ours, a subsequent problem reveals that these changes were ineffective or never made. Why this self-defeating pattern? What in this technology's culture of control might undermine experts' best intentions? What kind of problem is it to reduce operating risks? Following brief highlights of this industry's history over the last twenty years of accidents, near-accidents, and institutional changes, "Shouldering Risks" presents excerpts from interviews with some sixty experts about four relatively recent events at three U.S. plants. Drawing also on her earlier field studies at eleven plants in America and abroad, on industry documents, and others' research, Constance Perin identifies unacknowledged elements in this industry's culture of control - for example, control concepts for reactor design, construction, and regulation carry over to risk handling and event analysis, whose efficacy depends instead on recognizing and interpreting the significance of technical and contextual signals on daily display. Far more than the sum of its parts, this highly knowledge-dependent technology operates along an axis of meanings, not only along an axis of functions. A culture of control is, like any culture, an intricate system of claims about how to understand the world and act in it. Here, claims pivot around the dynamics of control theory and productivity based on particular assumptions about the relationships of humans to machines, models to reality, certainty to ambiguity, rationality to experience. These four events and accident analyses show that such assumptions can confound control and produce misleading meanings. "Shouldering Risks" reimagines a broader and deeper culture of control to reshape our understandings of the intellectual capital appropriate to designing, regulating, organizing, and managing this risky enterprise and, perhaps, other such technologies already here or to come.