Recent advances in biometrics, information forensics, and security have improved the recognition effectiveness of biometric systems. However, an ever-growing challenge is the vulnerability of such systems against presentation attacks, in which impostor users create synthetic samples from the original biometric information of a legitimate user and show them to the acquisition sensor seeking to authenticate themselves as legitimate users. Depending on the trait used by the biometric authentication, the attack types vary with the type of material used to build the synthetic samples. For instance, in facial biometric systems, an attempted attack is characterized by the type of material the impostor uses such as a photograph, a digital video, or a 3D mask with the facial information of a target user. In iris-based biometrics, presentation attacks can be accomplished with printout photographs or with contact lenses containing the iris patterns of a target user or even synthetic texture patterns. In fingerprint biometric systems, impostor users can deceive the authentication process using replicas of the fingerprint patterns built with synthetic materials such as latex, play-doh, silicone, among others. This research aimed at developing presentation attack detection (PAD) solutions whose objective is to detect attempted attacks considering different attack types, in each modality. The lines of investigation presented in this thesis aimed at devising and developing representations based on spatial, temporal and spectral information from noise signature, intrinsic properties of the biometric data (e.g., albedo, reflectance, and depth maps), and supervised feature learning techniques, taking into account different testing scenarios including cross-sensor, intra-, and inter-dataset scenarios. The main findings and contributions presented in this thesis include: the creation of a large and publicly available benchmark containing 17K videos of presentation attacks and bona-fide presentations simulations in a facial biometric system, whose collect were formally authorized by the Research Ethics Committee at Unicamp; the development of novel approaches to modeling and analysis of extrinsic properties of biometric samples related to artifacts added during the manufacturing of the synthetic samples and their capture by the acquisition sensor, whose results were superior to several approaches published in the literature that use traditional methods for image analysis (e.g., texture-based analysis); the investigation of an approach based on the analysis of intrinsic properties of faces, estimated from the information of shadows present on their surface; and the investigation of different approaches to automatically learning representations related to our problem, whose results were superior or competitive to state-of-the-art methods for the biometric modalities considered in this thesis. We also considered in this research the design of efficient neural networks with shallow architectures capable of learning characteristics related to our problem from small sets of data available to develop and evaluate PAD solutions.