We attempt to contribute to a general theory of requirements change from a goal-oriented and viewpoints-driven angle. To practitioners, this knowledge is relevant to anticipate changes in certain types of requirements, which may shorten the project's timeline, reduce costs, and increase product quality. Initially, we followed the common assumptions that what should be on a system is demanded by goals to achieve and what should not be on a system is demanded by goal states to avoid. However, requirements engineering of a diversity of systems (capacity and warehouse management, COTS PCs, and a Braille mouse) revealed that must requirements are predicted by goals to avoid (!) and won't requirements by goals to approach (!). Expectations about the positive or negative impact (valence) of requirements on goals played a moderating role. We unfold the gradual discovery of this "goals-to-requirements chiasm" (CHI-effect or χ-effect), claiming that variability in agreement to positive or negative requirements is predicted by goals of opposite polarity. We found that whether the χ-effect occurred or not, depended on the alignment of stakeholder viewpoints on goals and requirements. Comments from practitioners are included. Categories & Subject Descriptors: H.1.2 [Models and Principles]: User/Machine Systems-Human information processing; K.6.3 [Management of Computing and Information Systems]: Software Management-Software development. General Terms: Requirements Engineering, Human Factors, Theory. © 2006 Elsevier Inc. All rights reserved.