Wireless media access depending on packet size distribution over error-prone channels Ad Hoc Network is a decentralized type of network where wireless devices are allowed to discover each other and communicate in peer to peer fashion without involving central access points. In most ad hoc networks, nodes compete for access to shared wireless medium, often resulting in collision (interference). IEEE 802.11, a well-known standard, uses medium access control (MAC) protocol tosupport delivery of radio data packets for both ad hoc networks and infrastructure based network. But designing a Medium Access Control(MAC) protocol for ad hoc wireless networks is challenging, particularly when the protocol needs to achieve optimal performance both in terms of throughput and efficiency to deliver a packet. Error-prone channel has a significant impact on unsuccessful transmission probability which is often ignored by previous researches. Standard DCF (Distributed Coordination Function) operation of IEEE 802.11 enacted by binary exponential back-off (BEB) algorithm cannot differentiate collision from corruption and therefore sets forth a (time) separation between multiple nodes accessing the channel by (appropriately) adjusting contention window (CW) upon a failure. This leads to increased delay in error-prone network when nodes are not contending at all. Since packet corruption depends on bit error rate (BER) and length of packets, packet size can have significant impact on the throughput in error-prone environment. In this paper, we analyze effect of packet size in determining optimal CW to improve throughput and efficiency for error-prone networks. We propose a dynamic learning based scheme to adaptively select CW sub-range instead of whole selection range for different packet distribution. To validate our scheme extensive simulations have been done and simulation results show significant improvement in E2E delay performance.