Capacity sizing of business process integration (BPI) solutions deals with finding the lowest cost hardware configuration that can satisfy the performance requirements of a BPI solution being planned. Accurate sizing can help control the cost of a proposed BPI project while minimizing the risk of deploying a solution that will fail to meet the performance requirements later on. This paper proposes a model driven approach for predicting the performance of BPI solutions. The approach separates the application and middleware aspects of BPI solutions, and thus allows users to predict the performance of different BPI applications deployed on the same middleware. Using this approach, a sizing tool prototype for the Websphere InterChange Server is developed. Preliminary evaluation shows that the tool is able to recommend the hardware configuration with minimum cost while satisfying the performance objective of the applications. As a by-product, the sizes of the various thread pools of the middleware are also recommended by the sizing tool.