Reverberation in speech, caused by room reflections, is problematic especially for hearing impaired people, and therefore blind speech dereverberation is an important research area. The task is to remove reverberation from the output of a room, where the room as well as the speech input are unknown. In this project different approaches, based on higher-order statistics (HOS) and information theory, to the blind dereverberation problem, are analysed. The speech is assumed to consist of independent and identically distributed (iid) non-Gaussian samples, whereas the room output is assumed to consist of non iid Gaussian distributed samples. The method based on HOS maximizes the fourth order cumulant, referred to as kurtosis, because it is known that all Gaussian distributed signals have higher-order cumulants equal to zero, hence the output is then made as non-Gaussian as possible. For the information approach two methods are analysed: maximum and minimum entropy. The maximum entropy method maximizes the entropy through a nonlinear transformation, such that statistical dependency between samples is minimized. The entropy of the room output is minimized by the latter method, because Gaussian distributed signals lead to maximum entropy. The functionality of the three methods is analysed, tested, and compared on basis of the reduction of early and late reflections, convergence ability, and a real life application. Furthermore the complexity with respect to the computations needed by each method is calculated. Based on tests the project group presents adjustments to the methods which improve the performance on all fields for all methods. Furthermore the project group presents a novel algorithm based on the information theory methods for future work. The conclusion is that all three methods are able to significantly reduce the reverberation. The methods based on information theory seem to be superior to the method based on HOS.