The identification of catalytic residues of an enzyme is one of the most important steps towards understanding its biological roles and exploring its applications. Thus far, a range of catalytic residue prediction methods have been developed, which play an increasingly important role in complementing the experimental characterization of enzymatic functions. The available approaches can be split into two broad categories: i) similarity-based catalytic residue annotation and ii) de novo catalytic residue prediction. In this article, we review the existing research strategies, recently developed bioinformatics tools, and future perspectives in the topic of de novo catalytic residue prediction. In particular, we review the various residue properties that have been used to distinguish catalytic and non-catalytic residues. We also detail how these residue properties can be combined into a prediction system with the assistance of different statistical or machine learning methods. Since in many respects de novo prediction of catalytic residues is still in its infancy, in this review we also propose some hints that are likely to result in novel prediction methods or increased performance.