In this paper, we propose a new version of neural network algorithm to enhance the accuracy of detecting website phishing attacks. The proposed algorithm is presented to update weights on multi layers network. The updated algorithm initially started by collecting and storing data in phishing website dataset. We compare the performance of neural network with several machine learning algorithms. The feature selection method applied by the improved neural network algorithms is effective for characteristic capturing with reasonable results. The neural network techniques involve innovative phishing detection model to extract the significant phishing features and patterns. The evaluation of the proposed method accuracy. Accuracy measures the phishing websites correctly detected as trusted websites among all instances. The main conclusion gained from this research is the effectiveness of neural network in detecting website phishing attacks. In addition, the incremental method used for combination of different datasets provided a good insight. We can conclude that the accuracy rate is dependent to the feature counts and dataset size. In addition, the proposed model helps to avoid loss of cumulative knowledge over time and even with the change of characteristics of phishing websites with respect to the designers and developers of these websites.