Recognizing and summarizing persons' activities have proven to be effective for increasing self-awareness and enable to improve habits. Reading improves one's language skills and periodic relaxing improves one's health. Recognizing these activities and conveying the time spent would enable to ensure that users read and relax for an adequate time. Most previous attempts in activity recognition deduce mental activities by requiring expensive/bulky hardware or by monitoring behavior from the outside. Not all mental activities can, however, be recognized from the outside. If a person is sleeping, relaxing, or intensively thinks about a problem can hardly be differentiated by observing carried-out reactions. In contrast, we use simple wearable off-the-shelf single electrode brain computer interfaces. These devices have the potential to directly recognize user's mental activities. Through a study with 20 participants, we collect data for five representative activities. We describe the dataset collected and derive potential features. Using a Bayesian classifier we show that reading and relaxing can be recognized with 97% and 79% accuracy. We discuss how sensory tasks associated with different brain lobes can be classified using a single dry electrode BCI.