Abstract-In this paper, we design a device-free intruder detection and alarm system, named WiGarde by exploiting off-the-shelf Wi-Fi channel state information (CSI) to detect an intruder through door or window. WiGarde extracts the CSI amplitude information across MIMO antennas. We implemented WiGarde with commercial IEEE 802.11 NICs and evaluated its performance in two cluttered indoor environments. The system is robust and avoids false alarm occurrence, owing to our novel bad stream elimination algorithm. To extract the best feature, we design a new method to intercept the segment of the signal of intrusion based on wavelet analysis and dynamic time window based on Short-time Energy. We adopt Support Vector Machine (SVM) algorithm to classify human intrusion; our SVM algorithm could classify intrusion process with general walking through the area of interest. We compare WiGarde with the previous approaches; results show that our system outperforms the corresponding best CSI-based and RSSI-based in both of static and motion states. Our system gained high accuracy of 94.5% in a dynamic environment for intrusion through door or window.