Tracking process for manoeuvring targets is still the main problem in Kalman filtering [1-6]. The conventional variable dimension filter used for manoeuvring target tracking is not a practical filter because of the previous state vector estimates for several scans must be recomputed to avoid the filter divergence when a new target model is detected [1,3-5]. In this paper a real time adaptive variable dimension filter is proposed. This proposed filter is without the previous estimates to be recomputed. This proposed filter is also with a less filter divergence level and therefore more practical filter. Such a filter will be more appropriate to use when the target rather rarely experiences a new models. When the target more frequently experiences a new models it will be better to use the constant acceleration filter. This constant acceleration filter is actually a part of the proposed filter in this paper. This constant acceleration filter is also capable of
operating in real time with no filter divergence at the expense of slight increasing of the estimate error variance level.