In almost any sector of industry, implementation of smart maintenance can yield significant cost reductions or availability improvements. One of the elements of smart maintenance is the capability to predict upcoming failures and use that to apply just-intime maintenance. This predictive maintenance concept requires on the one hand appropriate monitoring strategies to collect information on the loads, usage or condition evolution of parts and systems. On the other hand, models and algorithms are required to translate this collected data into reliable predictions of the remaining useful life. Whereas the monitoring part is becoming more or less mature nowadays, the big challenge is in the processing of the data to achieve proper predictions. In this presentation, two approaches for this prognostics challenge will be discussed and compared. The first approach is based on models describing the physical degradation processes and the second approach is based on data analytics. The pros and cons of both approaches will be shown and various case studies will be used to demonstrate this, ranging from military vehicles and naval ships to wind turbines and helicopters.