The Weighted K Nearest Neighbors(WKNN) algorithm based on the off-line Received Signal Strength Indication(RSSI) fingerprint database has been studied intensively in the indoor positioning methods based on the received Wireless Fidelity(WiFi) signal. However, the specifications of the received RSSI fingerprint data, such as the high dimension and many invalid default RSSI values, have not been addressed in the existing WKNN algorithm, which is not good for improving its positioning accuracy. Aiming at the problems of the existing WKNN algorithm, the received RSSI values will be sorted in descending order, and the RSSI values larger than the preset threshold are selected to match with the off-line RSSI fingerprint database in the following steps. Then, the K value is determined on line adaptively by the statistics of the Euclidean distances. Finally, the Gaussian weights are updated by the means of the Euclidean distances. The experiment results show that the improved WKNN algorithm achieves more accurate positioning performance than the existing WKNN one.