One of the goals of smart environments is satisfying proactively its users' needs. To do so, it should understand what they are doing and what are their habits and preferences. In turn, this requires recognizing, at a lower level, single activities (e.g., reading, sitting, looking for books, …) and, at a higher one, how these activities are organized into daily routines. For these purposes we exploit the WoMan system. Using a process mining approach, it is able to incrementally learn a user's activities and daily routines as workflow models. Then, such models are exploited by WoGue (“Workflow Guesser”) system to recognize the current routine or activity of the user. The approach has been tested in a real-world setting, a smart office environment, equipped with a sensor network based on Arduino. We collected an annotated dataset of 45 days, and learned from it the workflow models of the user's daily routines and of his activities performed in the office. Results of some experiments show that how our approach is quite effective to learn and recognize activities and routines. Indeed, it achieves an average accuracy of 82% for activities and 98% for transitions among activities. Also the real-time recognition performance was tested, using sensor data coming from the smart office environment. In 82% of the cases the system correctly recognized the current activity and routine after just four events.