Loss Differentiation Algorithms (LDA) are currently used to determine the cause of packet losses with an aim of improving TCP performance over wireless networks. In this work, we propose a cross-layer solution based on two LDA in order to classify the loss origin on an 802.11 link and then to react consequently. The first LDA scheme, acting at the MAC layer, allows differentiating losses due to signal failure caused by displacement or by noise from other loss types. Moreover, in case of signal failure, it adapts the behavior of the MAC layer to avoid a costly end-to-end TCP resolution. The objective of the second LDA scheme, which acts at the TCP layer, is to distinguish a loss due to interferences from those due to congestions and to adapt consequently the TCP behavior. The efficiency of each LDA scheme and of the whole cross-layer solution are then demonstrated through simulations.