In a target tracking device, a state quantity estimation unit is configured to, every time a preset repetition period of a processing cycle elapses, estimate, for each of the one or more targets, a current state quantity based on at least either observation information of the one or more targets observed by a sensor or past state quantities of the one or more targets. A model selection unit is configured to, for each of the one or more targets, select one motion model from a plurality of predefined motion models, based on at least either states of the one or more targets or a state of the vehicle. An estimation selection unit is configured to, for each of the one or more targets, cause the state quantity estimation unit to estimate the state quantity of the target with the one motion model selected by the model selection unit.