Distributed systems have really gained ground in several applications domains. Recently, extracting valuable knowledge from large volumes of data in distributed setting is a major concern for most organizations. There is therefore need for architectures and models for efficient mining associations and connections from such data. Many models have been produced but no one model fits all scenarios. Massive data movement, high communication costs and ultimately slow response time are some major problems with existing models. This work therefore proposes an improved cost model for mining association rules from distributed databases.