Multimedia task composition involves automatic matching and selection of multimedia services across various devices in the pervasive environment. During the composition process, various QoS attributes are considered for selection of the service components implementing those services. However, the existing approaches about multimedia service composition do not consider user's perspective about utilizing services and devices. The automatic selection of service components keeps the user out of the whole process. The result is that the end-user may not always be satisfied with the final set of automatically chosen components and devices. In this paper, we propose a solution for multimedia task composition based on user preferences vis-a-vis device capabilities. User tasks are defined in terms of abstract services, which may also specify their requirements. Both of these criteria are considered simultaneously and a quantitative model for selection of service components and devices is presented