Each person using electrical devices leaves electricity fingerprints in the form of power consumption. These can be very useful for understanding the context of that person in, for instance, a smart office. A device that is highly correlated with the presence of a person in an office is the computer monitor; the correlation is in the range 83-96%. Therefore, it is useful to recognize from an aggregated power load the portion that is due to computer monitors. In this paper, we propose an event-based device recognition approach. After studying several predictors and features for device classification, we build a prototype for the classification. We evaluate the approach with actual power measurement of seven office monitors used by four workers in an office environment. Our experiments show that the approach is feasible and the per-day accuracy ranges in the range 69-80% for seven and five physical devices, respectively.