The goal of this thesis was to characterise a novel transmission detector in the context of signal prediction. This was to eliminate the need to collect a baseline signal for the device before treatment. This not only saves time, but, by independently generating the baseline signal, the process is less prone to missing errors. A simple analytical algorithm was designed and was found to be capable of detecting gross errors, however, it was shown not to be accurate enough to detect MLC position errors that could have a clinical effect on the delivery. MU check software was commissioned, however the fluence distribution it produced lacked the complexity for accurate signal prediction. A Monte Carlo model of a linac was built and validated then used to demonstrate that the detector could be modelled as two slabs of Perspex; the signal being proportional to the dose measured in the air between them. Two Monte Carlo models were then made using different systems, these were both evaluated by comparing predicted signals to measured signals for VMAT plans. Both models performed well and were capable of detecting leaf errors ~1mm; the merits of both are discussed with regard to error detection and ease of use.