The early 1990s of the 20. the century is marked as begining of the development of social networks. Home computers have become part of every household, bringing the number of users on social networks grew more and more. The development of technology, the home computers became capable of producing a large amount of data. The first social networks were in the form of a platform for communication, while today's social networks have achieved a great user capacity and production data, different semantics. The data can describe the personal preferences of the user, information about the education and the like. This situation has led to the problem of grouping large amounts of data with regard to the criteria that you want. Manually processing large amounts of data is time-consuming and exhaustively. Currently the most widely used algorithms for detecting communities in social networks have Edge, propagation betweeness Label, Fastgreedy, Walktrap, Spinglass, Infomap, the Leading eigenvector and Multilevel. The speed, the accuracy of the detected number of communities and modularity are the values that are observed for each algorithm. The algorithms are tested over the computer-generated networks and social network with social platform Facebook. For computer generated social networks it has been used Girvan-Newman's evaluation system and its generalization, Lancichinetti-Fortunato-Radicchi evaluation system. After conducting the test concluded that the algorithms of Multilevel and Walktrap the best algorithms according to their working properties. The worst performance was showed by Spinglass and Infomap algorithms due to lack of accuracy of the data, despite the short run time. On actual social network Spinglass shows a good performance, but I can carry out calculations of graphs with isolated nodes. An Edge betweeness algorithm requires most of the time, and the quality of the performance depend on the complexity of the graph, i.e., the greater the number of nodes, it is less accurate. Knowledge of the General behaviour of the algorithm against large data set allows scholars a quality way of work and a clever selection of who will help him when he met with a large data set, which from day to day, the development of technology is getting bigger.