In this paper, the possibility of using an emerging approach, namely device-free sensing (DFS) technology, for classification of human activities is investigated. To fully evaluate this approach, several samples have been collected in an outdoor open-field environment. Using the collected data along with a classifier, a high-order cumulant (HOC) based feature extraction algorithm is investigated. To demonstrate the improvement of using this algorithm, the classical approach that is based on received-signal strength (RSS) is chosen as a benchmark. The experiment results demonstrated that the classification accuracy of the proposed algorithm is better than the classical approach by at least 15%. In addition, the reliability of the presented approach due to variation of training samples and signal-to-noise ratio (SNR) are also carefully tested using experimentally recorded samples, so that a good reliability can be ensured.