This paper deals with the problem of predicting the popularity of online news.With development of technology and the expansion of the Internet, users can access "sea" of content at any time. Therefore, the question is how to stimulate interest among other users of the internet for content that we publish. In this paper I propose the use of the knowledge of neural networks, which would analyze the content of article before it is published online. Neural networks are used in a wide range of areas, like the problems of classification, prediction, and for pattern recognition in data. By analyzing 58 elements (eg. keywords, digital media content ...) neural network would classified online news into category popular or unpopular. Data used to conduct this study were taken from the database Online News Popularity Data Set, which can be found at the website of University of California Irwin Machine Learning Repository. The database has of over 39,000 analyzed articles. The data was collected by Kelwin Fernandes, Pedro Vinagre and Paulo Cortez. During this research 40 architecture of neural, networks were tested, during the process of building architecture of neural networks the parameters (activation function, error, sample layout, etc.) of the research have been changed in order to improve the model. The network with the highest classification accuracy was selected for the best model. Below you will find explained in detail the matrix of confusion of the best model. Also, an analysis of the sensitivity was conducted, to determine which elements of article has greatest impact on its popularity.