Important sources of railway noise are caused by the wheel/rail interaction, including both rolling and impact noise. Although a considerable amount of research has been carried out into rolling noise, far fewer investigations have been carried out into railway impact noise. This occurs due to discrete surface discontinuities such as wheel flats, rail joints and turnouts and leads to an increase in the noise level. Whereas rolling noise can be predicted in the frequency domain, time-domain methods are required to investigate impact noise. Previously a hybrid method has been developed to predict the overall and third-octave band spectra of sound pressure levels. However, it cannot provide time histories of the sound pressure signal required for obtaining the peak sound pressure. These gaps led to the development of a time-domain model for predicting railway impact noise in this thesis. This thesis presents the development of a railway impact noise model for predicting the sound pressure time histories due to impact. The wheel/rail interaction model includes the non-linearity of the contact and potential contact loss. The effect of wheel rotation is also considered. To predict the radiated noise, a three-dimensional boundary element model is used for the wheel and an equivalent source model for the track. A Fourier transform is used to convert these results back into the time domain. To verify the model, results were predicted for a random roughness input, and these were compared with the TWINS model in terms of frequency spectra of the contact forces, wheel and rail vibration and radiated sound. The predictions agreed well with each other. Investigation of the response to a typical discontinuity representative of a crossing indicated that when the wheelset’s flexible modes are included in the prediction, the peak load, and consequently the peak sound pressure, are affected. Including the effect of wheel rotation showed an insignificant effect on the vertical vibration and contact force. However, it notably affected the lateral wheel response and radiated noise. The radiated noise was determined at a position opposite the wheel in a frame of reference moving with the wheel. The contributions of each component to the noise indicated that the wheel had the highest contribution to the impact noise. As is commonly found for rolling noise, the wheel contributes the most at high frequency. However, it was found that the wheel also dominated the noise in the mid-frequency region, especially due to the umbrella mode of the wheel (n=0, m=0). The relationship between the peak sound pressure level and dip heights (𝐻) was investigated, indicating a relationship of about 20log10𝐻. Similarly, the relationship between the peak sound pressure level (Unweighted and C-weighted) and speed (𝑉) showed a relationship of about 20log10𝑉. Similar findings are obtained for the overall A-weighted level when considering the impact noise within a certain travelling period. When observing the impact noise in terms of overall A-weighted level for a certain travelling distance (such that the time period varies with speed), it showed a relationship of about 30log10𝑉, similar to the rolling noise. Lastly, the effect of the loss of contact was investigated and found to have an insignificant influence on the impact noise. However, the loss of contact influenced the rail radiation. Finally, a wheel-on-disc rig (originally a pin-on-disc machine) was designed to replicate the rolling and impact noise in a controlled environment. The experiment successfully observed the characteristic of rolling and impact noise in terms of the speed dependence of the overall A-weighted level. However, due to the high background noise presented during the experiment, it was not possible to observe the characteristic of the rolling noise. The feasibility of using a laser Doppler vibrometer was investigated. This showed that it has the potential for contactless measurements of the rotating wheel. Still, it requires further processing to improve the signal quality, which is reduced by typical speckle noise and dropout issues.