The problem of object localization in Wireless Sensor Networks is dealt with a Support Vector Machine classifier exploiting the Received Signal Strength Indicator measured by WSN nodes. The SVM is trained once and offline, so the real-time functionality during the test phase is guaranteed. Unlike other methods devoted to nodes localization, the proposed approach solves the problem of localization and tracking of an unknown object (e.g. human beings) that does not belong to the WSN infrastructure. A preliminary result from the assessment of the feasibility of the SVM-based localization/tracking method is presented and discussed.