|Unmanned Aerial Vehicles (UAVs) are an increasing presence around the world; however, they can pose a threat to secure facilities. Many UAV mitigation techniques require accurate knowledge of UAV states to successfully intercept an adversarial UAV, but access to UAV on-board sensors may not be possible. One potential solution to this problem is to estimate UAV states using only radar measurements. This scenario is examined in simulation and with real world data. A discrete Extended Kalman Filter (EKF) with a constant acceleration dynamic model provides a baseline estimation performance of simulated UAV maneuvers and is shown to have consistent error in state estimates during high dynamic maneuvers. The simulated UAV maneuvers are then modelled as Hidden Markov Models (HMMs). HMMs are utilized to perform real time classiﬁcation of maneuvers and to provide acceleration and jerk estimates of the UAV through the use of a Gaussian Mixture Regression. HMM classiﬁcation of simulated maneuvers results in high accuracy classiﬁcation during UAV ﬂight. The HMM acceleration and jerk estimates are then incorporated into a state estimation framework as inputs to the ﬁlter’s dynamic model. This new system is known as the EKF+HMM. When estimating high dynamic maneuvers, the EKF+HMM performs better than the baseline EKF, while performing at similar levels when estimating low dynamic maneuvers. HMM classiﬁcation and the EKF+HMM are also tested on a real world data set of maneuvers performed by a Tarot X8 Octacopter. HMMs were trained for each maneuver, using experimental data or simulated data. HMM classiﬁcation was successful using both types of HMMs, although models trained with experimental data performed better. The EKF+HMM was also tested on the real-world data set and performed worse than the EKF when using simulation data trained HMMs and at the same level as the EKF when using HMMs trained with experimental data.