The description of protein 3D structure usually focuses on the repetitive local folds (alpha-helices and beta-sheets). The remaining class, sometimes called unordered region, has often been considered as random (one also calls it ‘random coil'). However, some interesting local folds are also highly recurrent and definitely more structured than a real random region. One of such particularly interesting motif is tight turn; this latter is characterized by few residues (3 to 5) and by the reversal of the polypeptide chain. This reversal of the protein backbone plays an important role in the topology of peptides or proteins, and can be directly associated to biological functions. Tight turns are characterized by precise dihedral angle values of their central residues and short distance between their extremities (implying thus the reversal). Among all tight turns, gamma-turns are the most important and studied class. They are made of 4 residues and account for 25% of all residues in proteins. According to the values of the dihedral angles of the two central residues, several types are defined. Some specific types have been shown to be involved in specific biological functions. Like for secondary structure, turn prediction has greatly evolved these last decades. Various methods have been used for predicting their presence such as statistical ones, artificial neural networks (ANN) or very recently support vector machines (SVM). In the 1970's, the first methods used only the amino-acid sequence to predict the presence of turns. Some great improvements were made by incorporating evolutionary information and by using predicted “classical” (i.e. alpha-helices, beta-sheets and coil) secondary structure. Some of the latest methods go further and predict not only the presence or absence of turns but also their type. In this paper, we briefly review the discovery and evolution of tight turn definition; we then present the different prediction methods and highlight the importance of the statistical evaluation of the results.