Limited by the sampling capacity of the mobile devices, many real-time indoor location systems have such problems as low accuracy, large variance, and non-smooth movement of the estimated position. A new positioning algorithm and a new processing method for sampled data are proposed. Firstly, a positioning algorithm is designed based on the cluster-based nearest neighbour or probability. Secondly, a weighted average method with sliding window is used to process the sampled data as to overcome the mobile devices’ weak capability of signal sampling. Experimental results show that, for the general mobile devices, the accuracy of indoor position estimation increases from 56.5% to 76.6% for a 2-meter precision, and from 77.4% to 90.9% for a 3-meter precision. Therefore, the proposed methods can significantly and stably improve the positioning accuracy.