Smart environments rely on artificial intelligence techniques to make sense of the sensor data that is collected in the envi-ronment and to use the information for data analysis, predic-tion, and event automation. In this paper we discuss an im-portant smart environment technology — resident activity recognition. This technology is beneficial for health moni-toring of a smart environment resident but accurate recogni-tion is difficult for real-world situations. We describe our approach to activity recognition and discuss how incorporat-ing temporal reasoning improves the accuracy of our algo-rithms. We validate our algorithm on real sensor data col-lecting in our smart apartment testbed.