Electron beam welding (EBW) is a joining process that has been widely applied in many modern industrial sectors. However, in order to achieve a satisfactory welding quality for a given material and configuration, a trial-and-error approach is usually adopted before moving to the final production. This procedure is often wasteful, time consuming and expensive when the raw material is at high cost, and greatly relies on the operators’ personal experience. To enable a ‘smarter’ welding process and reduce the inconsistent human factor, this PhD study is to develop a novel method based on statistic modelling, numerical modelling and artificial neural networks to predict the weld profile, which is the main criterion for assessing the welding quality. The models are set up with electron beam characteristics collected through a 4-slits technology to determine the actual focal spot size and power density, therefore the uncertainty caused by beam variation can be reduced. Multi-influences caused by electron beam, machine parameters and process environment are considered, and the predictions cover a wide range of linear beam power ranging from 86 J/mm to 324 J/mm. Finally, a novel simulation tool for predicting electron beam weld shape has been developed with assistance of a 4-slits beam probing technology to reduce the amount of manual work traditionally needed to achieve high-efficiency and high-quality welding joints. Validated by experimental results, the model is able to predict the weld profile with high accuracy and reliability for both partially and fully penetrated welding situations. By combining the numerical model and artificial intelligence, a weld-profile prediction system is to be integrated in current EB welding machines to allow a less-experienced operator to achieve high welding quality.