Electric vehicles emit less noise at low speeds compared to internal combustion engine vehicles. Thus, auditory cues necessary for other road users to detect an approaching electric vehicle is less than for internal combustion engine vehicles. Consequently, warning sounds are emitted from electric vehicles to maintain the same level of auditory cues. It is of great interest to maximize the detectability of these warning sounds while keeping the perceived annoyance as low as possible. The current procedure for determining detectability of sounds is to conduct experiments and listening tests which are very comprehensive procedures. In this project it is investigated if a computational partial loudness model can be used for predicting detectability as it would improve the warning sound design/test process. A partial loudness model has been implemented in MATLAB and different listening tests have been carried out to obtain subjective detectability and perceived annoyance data for different warning sounds in different noise conditions. Using the implemented model on the experiment data it was shown that the model performed well on stationary sounds and performed well across noise conditions. A detection threshold of 27 phon was obtained for stationary warning sounds. However, the partial loudness model could not appropriately predict neither detectability nor perceived annoyance across all tested warning sounds and noise stimuli. Improvements specifically concerning modelling the temporal effects should be made before the partial loudness model can be used as a universal model for predicting detectability for all sound/noise stimuli. Furthermore, it should be investigated how other sound quality metrics correlate with perceived annoyance in order to make a universal annoyance metric specifically made for EV warning sounds.