The present status is given for classification and datafusion within the project Multisensor Target Seeker Demonstrator. We have studied the ability of a target seeker to recognise previously detected combat vehicles. A newly developed database stores various scenarios containing six different combat vehicles on a grass field. The targets have been observed at constant 300 m height, approached from 8 km distance down to 500 m. Three test cases cover different sets of target range, aspect angles and articulation of its turret. The combat vehicles, seen in IR, have been segmented using adaptive thresholding. From the resulting binary images various features have been extracted, such as area/perimeter2, with a special emphasis on range invariant properties. Using linear discriminant analysis these features are combined into new and fewer features that maximise the class separability. As regards radar, the complete range profile makes a feature vector. Using Singular Value Decomposition representative feature profiles have been extracted from optimally sized groups of range profiles observed at 0.025 degree step in elevation angle. The so chosen features serve as training data for the classification algorithm, RPROP, an artificial neural net based on resilient backpropagation. The results show that targets are classified correctly more successfully using combined IR and radar, given the uncertainty in sensor data. The recognition capability is degraded for articulated targets having their turret turned away from its forward pointing position. For radar, in particular, it is important that training data are accumulated from observations with small steps in target elevation angles (short ranges). Ground truth measurements using FOI´s radar "Arken" have partly verified the clutter model from Georgia Institute of Technology (GIT) used for the radar simulations.