Abstract: Multiple expert decision combination strategies have been used extensively in designing very powerful classifiers for various image processing tasks. These approaches are generally very successful in enhancing the recognition performance of a system, but tend to be costly in terms of implementation and execution, making their application in real time processing environments difficult. This paper investigates the implications in terms of processing speeds and other implementability issues in relation to the incorporation of these multiple expert decision combination approaches in system design. It is demonstrated that selection of a particular multiple expert approach for a particular task domain is influenced by both the achievable recognition performance and the overall execution speed in terms of system throughput. A performance-cost profile has also been proposed to visualise and select the optimal decision combination approach for a specific task domain.