In order to read proficiently we must be able to tackle printed words that have not been explicitly taught. The process of rearranging previously learned linguistic units into novel patterns is referred to as recombinative generalisation. In the case of reading, sub-word units (e.g., onsets and rimes, phonemes, syllables) of known words are recombined to form new words that can be pronounced by reference to letter–sound correspondences. Matching-to-sample (MTS) procedures have proven to be effective in promoting recombinative generalisation. The aim of the current thesis was to further develop a MTS training protocol to facilitate recombinative generalisation. Additionally, this research aimed to extend the use of such MTS procedures by exploring ways in which the training protocol could be applied to help support recognition of the wide assortment of words encountered by beginning readers of English (e.g., monosyllabic and disyllabic words, consonant blend words, regular and exception words). Seven experiments were undertaken with literate adult participants to thoroughly test the effectiveness of the new MTS training protocol. A novel (invented) script was constructed for use with the adult participants. The findings from these experiments are reported in the first part of the thesis. To begin the research with the adults, modifications were made to MTS procedures developed by Mueller, Olmi, and Saunders (2000) and Keaveney (2005). The adapted MTS training protocol included conditional discrimination training, explicit symmetry training, and mixed symmetry testing for onset and rime sound–symbol relations. Experiment 1 was designed to test the efficacy of the revised protocol. All participants successfully completed the new training procedures and could accurately identify the novel recombined onset–rime and onset–rime–rime test words. To isolate the key training components within the protocol necessary for recombinative generalisation to occur, each of the training components was removed separately from the procedure in Experiment 2. The data from Experiment 2 clearly demonstrated how vital conditional discrimination training is to facilitating the recombinative generalisation effect. For Experiment 3, a new across-unit recombinative generalisation test was incorporated within the protocol. Recognition of recombined words composed from untrained onsets and rimes (derived from the explicitly taught onsets and rimes) was examined. In a bid to improve performance on the across-unit test, an intermediate across-unit test was added to the protocol in Experiment 4. Overall, the data from Experiments 3 and 4 indicated that the MTS protocol did appear to be effective in promoting across-unit recombinative generalisation. Likewise, in Experiment 5, the protocol was found to be effective in facilitating recognition of words containing consonant blends. In Experiment 6, variations were made to the grain size of the training unit employed to train the sound–symbol relations. Participants completed training with onsets and rimes, phonemes, or syllables (i.e., whole CVC words). The size of the training unit employed was found to have a significant effect on accuracy in recognising the recombined test words. Although there was no difference in the recombinative generalisation performance of onset and rime and phoneme trained participants, both groups of participants identified significantly more recombined words than the syllable trained participants. The final adult experiment was designed to examine if the protocol could be used to facilitate recognition of regular (rime consistent) and exception (rime inconsistent) words. One of the main findings to emerge from this experiment was that MTS training at only one grain size (i.e., onsets and rimes or phonemes or syllables) appeared to be insufficient to enable participants to tackle both the regular and exception words. For the second part of the thesis, the protocol was used as a remedial tool within a classroom setting with seven children with mild learning disabilities. The aim of this phase of the research was to see if the protocol could help improve the children’s word recognition and naming skills for actual words. Baseline reading assessments were undertaken prior to the start of training to measure key reading related skills such as phonological awareness and letter–sound knowledge. A subsequent study was conducted to determine if there was an optimal sized training unit for each child (e.g., onsets and rimes, phonemes, or syllables) that appeared to best facilitate recombinative generalisation. All of the children demonstrated gains in word naming and recognition following completion of the MTS training procedures, most notably, following onset and rime training. To try and strengthen the effectiveness of the procedure, a stimulus equivalence training component, in which the children learned to relate printed words, spoken words, and corresponding pictures, was incorporated into the protocol. The stimulus equivalence component was used in a novel way to train recognition of regular and exception words. Again, encouraging results were obtained, with all of the children showing gains in their ability to read aloud and recognise trained and untrained (recombined) words. Implications of both the adult and child research are discussed in the final chapter of the thesis. There is a particular focus on how there is much to be potentially gained from integrating behavioural methods with cognitive theories of reading when the aim is to develop effective aids to teach components of reading.