The aim of this research is to improve routing fault tolerance in Mobile Ad hoc Networks (MANETs) by optimising mUltipath routing in a well-studied reactive and single path routing protocol known as Ad hoc On-demand Distance Vector (AODV). The research also aims to prove the effect of varying waiting time of Route Reply (RREP) procedure and utilising the concept of efficient routes on the performance of multipath extensions to AODV. Two novel multipath routing approaches are developed in this thesis as new extensions to AODV to optimise routing overhead by improving Route Discovery Process (RDP) and Route Maintenance Process (RMP) of multipath AODV. The first approach is a Iinkdisjoint multipath extension called 'Thresho)d efficient Routes in multipath AODV' (TRAODV) that optimises routing packets ~verhead by improving the RDP of AODV which is achieved by detecting the waiting time required for RREP procedure to receive a threshold number of efficient routes. The second approach is also a link-disjoint mUltipath extension called 'On-demand Route maintenance in Multipath AoDv' (ORMAD) which is an extension to TRAODV that optimises routing packets and delay overhead by improving the RMP of TRAODV. ORMAD applies the concepts of threshold waiting time and efficient routes to both phases RDP and RMP. It also applies RMP only to efficient routes which are selected in the RDP and when a route fails, it invokes a local repair procedure between upstream and downstream nodes of the broken link. This mechanism produces a set of alternative subroutes with less number of hops which enhances route efficiency and consequently minimises the routing overhead. TRAODV and ORMAD are implemented and evaluated against two existing multipath extensions to,AODV protocol and two traditional multipath protocols. The existing extensions to AODV used in the evaluation are a well-known protocol called Ad hoc On-demand Multipath Distance Vector (AOMDV) and a recent extension called Multiple Route AODV (MRAODV) protocol which is extended in this thesis to the new approach TRAODV while the traditional multipath protocols used in the evaluation are Dynamic Source Routing (DSR) and Temporally Ordered Routing Algorithm (TORA). Protocols are implemented using NS2 and evaluated under the same simulation environment in terms of four performance metrics; packet delivery fraction, average end-to-end delay, routing packets overhead, and throughput. Simulation results of TRAODV evaluation show that the average number of routes stored in a routing table of MRAODV protocol is always larger than the average number of routes in TRAODV. Simulation results show that TRAODV reduces the overall routing packets overhead compared to both extensions AOMDV and MRAODV, especially for large network size and high mobility. A vital drawback of TRAODV is that its performance is reduced compared to AOMDV and MRAODV in terms of average end-to-end delay. Additionally, TORA still outperforms TRAODV and the other extensions to AODV in terms of routing packets overhead. In order to overcome the drawbacks of TRAODV, ORMAD is developed by improving the RDP of TRAODV. The performance of ORMAD is evaluated against RREP waiting time using the idea of utilising the efficient routes in both phases RDP and RMP. Simulation results of ORMAD show that the performance is affected by varying the two RREP waiting times of both RDP and RMP in different scenarios. As shown by the simulation results, applying the short and long waiting times in both phases tends to less performance in terms of routing packets overhead while applying the moderate waiting times tends to better performance. ORMAD enhances routing packets overhead and the average end-to-end delay compared to TRAODV, especially in high mobility scenarios. ORMAD has the closest performance to TORA protocol in terms of routing packets overhead compared to ~M~a~M~OW . Relevant concepts are formalised for ORMAD approach and conducted as an analytical model in this thesis involving the\vhole process of multipath routing in AODV extensions. ORMAD analytical model describes how the two phases RDP and RMP interact with each other with regard to two performance metrics; total number of detected routes and Route Efficiency.