Dedicated bird radars have been used in ornithological studies for many years. This techniques has the advantage that it provides continuous data on a large scale. However, there are also several restrictions to this technique: the recorded radar data have a low taxonomic resolution and radars also records objects other than birds (e.g. sea surface, ships, rain). All unwanted detections are being referred to as clutter. The goal of this study is to develop a reliable filter, based on the differences in target characteristics as recorded by the radar, to post-process the vertical radar data which removes as much clutter from the database as possible. This will result in a more accurate bird flux and therefore an improved outcome of the bird collision model. The model tests showed very high scores for the criteria accuracy, sensitivity and specificity. The model precision is a lower in one of the two tests. This is caused by a relatively high number of false positives in the model results. This will be improved in the future by including variables in the decision tree analysis which are linked to the bird track level, instead of only using the variables recorded by the radar which describe the single targets, as was the case in the current model.