In consideration of the direct influence of Received Signal Strength Indicator(RSSI) fluctuation on the target localization accuracy in wireless sensor networks (WSN), the basic principle of target localization using Least Square Support Vector Regression(LSSVR) is discussed. Then, the characteristics of LSSVR modeling are analyzed for given and variable detection sensors, respectively. Furthermore, a method for Target Localization based on Adaptive LSSVR Modeling (TL-AML) in WSN is proposed. By considering localization accuracy and real-time performance comprehensively, LSSVR models are built for locating target at initial time, and at follow-up time it is used to decide whether new models need to be built or not according to the inclusion relation between detection nodes and sensor nodes. The performance of TL-AML is verified based on CC2430 WSN experiment platform. Results show that the Root Mean Square Error (RMSE) of target localization based on TL-AML has reduced by 34%~37% and 60%~65% as compared with those of MLE and LSE,respectively. When modeling parameters are taken in reasonable value ranges, the localization accuracy of TL-AML is improved evidently compared with MLE and LSE. Moreover,the consuming time of TL-AML is 0.2~0.4 s,If LSSVR modeling is needed. Otherwise, the consuming time is only about 0.04 s. The results indicate that TL-AML method can weaken the influence of RSSI fluctuation on the accuracy of target localization and has good real-time target localization accuracy.