The continued escalation of manageable, wireless enabled devices with immense storage capacity and powerful CPUs are making the wide spread use of mobile databases a reality. Mobile devices are increasingly used for database driven applications such as product inventory tracking, customer relationship management (CRM), sales order entry etc. In some of these applications Location Dependence Data (LDD) are required. The applications which use LDD are called Location Dependent Information Services (LDIS). These applications have changed the way mobile applications access and manage data. Instead of storing data in a central database, data is being moved closer to applications to increase effectiveness and independence. This trend leads to many interesting problems in mobile database research and cache replacement policies. In mobile database system caching is the effective way to improve the performance since new query can be partially executed locally. The desired caching can be achieved by convincingly accurate prediction of data items for the present and future query processing. It is important to take into consideration the location and movement direction of mobile clients while performing cache replacement. Due to cache size limitations, the choice of cache replacement techniques used to find a suitable subset of items for eviction from cache becomes important. In this paper we propose a Markov Model based Cache Replacement Policy (MMCRP) for mobile database environment. Our method predicts the new data item to be fetched by searching second and/or first order transaction probability matrix (TPM) of Markov Model for valid scope, access frequency, data distance and data size. The implementation of these policies has been done using java. Simulation results for query interval, client speed and cache size show that the MMCRP performs significantly better than Least Recent Used (LRU), Furthest Away Replacement (FAR) and PPRRP.