Intrusion detection systems (IDS) have become an essential issue for computer networks security since each one is vulnerable for violation. This paper presents a neural network based implementation of an intrusion detection system to detect network based attacks. The key idea is to extract the most useful set of features from the packets traversing through the network and utilize them to describe users behavior. These selected features will be used an input features to train a designed neural network architecture to build a classifier that can recognize anomalies and known intrusions. Using a benchmark data set from a KDD (Knowledge Discovery and Data Mining), the designed system was able to correctly detect 99.8% of unusual network activity with a maximum of 5.4% false alarms. In addition, the system was 98.6% accurate in detecting different intrusion types.