Existing anomaly detection systems do not reliably produce accurate severity ratings for detected network events, which results in network operators wasting a large amount of time and effort in investigating false alarms. This project investigates the use of data fusion to combine evidence from multiple anomaly detection methods to produce a consistent and accurate representation of the severity of a network event. Four new detection methods were added to Netevmon, a network anomaly detection framework, and ground truth was collected from a latency training dataset to calculate the set of probabilities required for each of the five data fusion methods chosen for testing. The evaluation was performed against a second test dataset containing manually assigned severity scores for each event and the significance ratings produced by the fusion methods were compared against the assigned severity score to determine the accuracy of each data fusion method. The results of the evaluation showed that none of the data fusion methods achieved a desirable level of accuracy for practical deployment. However, Dempster-Shafer was the most promising of the fusion methods investigated due to correctly classifying more significant events than the other methods, albeit with a slightly higher false alarm rate. We conclude by suggesting some possible options for improving the accuracy of Dempster-Shafer that could be investigated as part of future work.