Computer-mediated collaboration, a dominant mode of organizational communication particularly in dispersed and multinational organizations, introduces unique opportunities but also new problems. One of these problems is the higher risk of misunderstandings, which is more likely to occur in computer-mediated teamwork than in face-to-face teams (Cramton, 2001). Moreover, it may be particularly acute when distributed workers come from different functional backgrounds, holding different perspectives (Dougherty, 1992; Powell et al., 2004). Dispersed collaborations are also likely to suffer problems of culture since collaborators are embedded in a different local work setting with its own rules, language, histories, and myths (Armstrong & Cole, 2002). Current theories of communication suggest that misunderstanding may be reduced by contextualization, i.e., providing contextual information to explain a core message. However, a central hypothesize in the current research, is that contextualization is beneficial in some situations but not in others. Treating contextualization as a form of adaptive behavior, the research model aims to understand its contingent impact on performance in collaborative tasks. The motivation for contextualization is explained, arguing that it can be predicted by the extent to which the perspectives of the collaborators are different or shared: a difference of perspectives between collaborators motivates them to contextualize in order to increase mutual understanding and thereby increase performance. Computer support should also motivate communicators to contextualize by making it easier for them to do so. A controlled experiment tested these relationships in a collaborative machine-assembly task performed by dyads. The collaborators' perspectives and the level of computer support were manipulated, and contextualization behavior, mutual understanding and performance were measured. Results show that contextualization is effective only for dyads with different perspectives and may be detrimental when perspectives are similar. When computer support is available, users may contextualize even if it is counterproductive. Therefore, computer-mediated collaboration should be designed to ensure only effective contextualization. Some potential practical implications for collaborative systems are offered.