Steganography is the art of providing a secret communication channel for the transmission of covert information. At the same time, it is possible that it can be used by cyber criminals to conceal their works. This potential illegal use of steganography is the basis for the objectives in this thesis. This thesis initially reviews the possible flaws in current implementations of steganalysis. By using images from different camera types, this thesis confirms the expectation that the steganalysis performance is significantly affected by the differences in image sources. In this thesis we prove that image compression in a steganalysis process has an impact on the steganalysis performance, as claimed in the literature. A review of currently available steganalysis techniques, along with a proposal to overcome the said problems is also presented in this thesis. We propose a new technique for steganography that is based on conditional probability statistics. This new technique works on 72 features (conditional probability features) extracted for each image for the purpose of classification. Through experiments based on standard benchmarks, comparable classification accuracies have been achieved by this new approach. Furthermore, these new features demonstrated good performance when applied to image forensic tasks. Applied to images from four digital cameras, these new features are able to classify test images according to their sources with accuracy rates of 99. 5% and 91. 5% in both inter-camera and intra-camera model cases, respectively.