In this paper, we introduce BINet, a neural network architecture forreal-time multi-perspective anomaly detection in business process event logs.BINet is designed to handle both the control flow and the data perspective of abusiness process. Additionally, we propose a set of heuristics for setting thethreshold of an anomaly detection algorithm automatically. We demonstrate thatBINet can be used to detect anomalies in event logs not only on a case levelbut also on event attribute level. Finally, we demonstrate that a simple set ofrules can be used to utilize the output of BINet for anomaly classification. Wecompare BINet to eight other state-of-the-art anomaly detection algorithms andevaluate their performance on an elaborate data corpus of 29 synthetic and 15real-life event logs. BINet outperforms all other methods both on the syntheticas well as on the real-life datasets.