Service virtualisation is a method to create virtual service models that can mimic interaction behaviors between a system under test and the target system. With service virtualisation, the development team can get access to the production-like conditions whenever and however many times they need, enabling frequent and comprehensive testing. Previous techniques for service virtualisation have relied on explicitly modelling the target services by a service expert and require detailed knowledge of message protocol and structure. However, neither of these are necessarily available. In this thesis, we introduce our novel opaque response generation approach. This approach enables services to be virtualised automatically without any expert knowledge or documentation of system protocol and interaction behaviours. Given a collection of interactions exchanged between a system under test and a target real service, our approach can 1) organise the same type of interactions into the same cluster and derive a cluster prototype for each cluster; 2) search a given incoming request for its the most similar request in the interaction library; 3) learn knowledge from the incoming request and the recorded interaction; and 4) generate a response. A framework and proof-of-concept implementation of our opaque response generation approach is described. Experimental results show our opaque response generation approach is able to automatically generate accurate responses in real time with an accuracy rate over 99% on average.