From the user's point of view, Earthquake Early Warning Systems (EEWSs) have a number of applications that need to be viewed in detail for better utilisation of the “lead time” and the usefulness of the associated information . From the seismological point of view, the most important information is of two types: (1) the amount of lead time, the period from the moment the end-user receives the warning until the moment of arrival of larger S waves of significant importance or of exceeding a threshold value of a parameter characteristic of the seismic motion and (2) the reliability of the information transmitted. Missing events and false alarms may be critical or not to the type of “facilities/equipment” we are trying to protect, depending of the consequences. And to be more confident of the predictions, the lead time becomes shorter because the number of stations required increases. To check the level of possible lead time for the Portuguese industrial complex of Sines, we used the available procedures (front and on-site detection for SS and SP wave arrivals) and published the results obtained with the present configuration of the station network and with a hypothetical station configuration. Monte-Carlo simulation was used for the epicentre location within the most critical seismic source zones. The level of reliability and useful lead time ideal for different operators may be quite different, depending on the type of equipment under analysis. Therefore, the optimum balance between reliability and lead time may vary significantly between end-users, and some may even be interested in more than one option. In this article we study the effect of these problems on the industrial infrastructures, a group of installations where EEWSs may have a tremendous impact. Lead times, false and missing events are analysed from the end-users' viewpoint. We applied a simplified and preliminary cost-benefit analysis of using EEWSs at an industrial site and concluded that it is worth doing for more frequent events, likely to cause some damage, but refinement of the modelling parameters deserves to be continued.