Abstract: To reduce the negative influence of the complex indoor environment on WiFi fingerprinting, the paper proposes a support vector machines (SVM)-based WiFi fingerprinting algorithm which combines SVM classification and regression for more accurate location estimation. For smartphone based pedestrian dead reckoning (PDR), the paper detects the steps by recognizing the state transitions during human walking using real-time acceleration data. To reduce the measurement noise and the accumulation of positioning errors, the paper proposes a pre-processing algorithm on the original acceleration data and determines the state transition parameters dynamically according to the real time acceleration data. Based on the SVM-based WiFi fingerprinting and the enhanced PDR, the paper uses Kalman filtering to fuse them for more accurate and more stable positioning results. Experiments show that the proposed algorithms are quite effective.