Abstract-Activity recognition has been widely studied in ubiquitous computing since it can be used in several application domains, such as fall detection and gesture recognition. Initially, works in this area were based on research-only devices (bodyworn sensors). However, with advances in mobile computing, current research focuses on mobile devices, mainly, smartphones. These devices provide Internet access, processing, and various sensors, such as accelerometer and gyroscope, which are useful resources for activity recognition. Therefore, many studies use smartphones as data source. Additionally, some works have already considered the use of wristbands and specially-designed watches, but fewer investigate the latest marketable wearable devices, such as smartwatches, which are less intrusive and can provide new opportunities to complement smartphone data. Moreover, for the best of our knowledge, no previous work experimentally evaluates the impact caused by the combination of sensor data from smartwatches and smartphones on the accuracy of activity recognition approaches. Therefore, the main goal of this experimental evaluation is to compare the use of data from smartphones as well as the combination of data from smartphones and smartwatches for activity recognition. We evidenced that the use of smartphone and smartwatch data combined can increase the accuracy of activity recognition.