In the last 20 years the advent of new technologies, such as high-throughput screening (HTS) and combinatorial chemistry, has produced new tools for the discovery of biologically active molecules. In the past decade, fragment-based drug discovery has emerged as a more rational and focused approach that concentrates on the quality, rather than the quantity, of hits and leads. The principles behind this strategy are different from those that represented the basis of conventional HTS. The starting point of this approach is always a small chemical entity (typically MW 150-200), a fragment, with low affinity for the selected target. Fragments should satisfy key features such as diversity, reduced structural complexity, aqueous solubility and availability. Because of their small size, they occupy a smaller region of chemical space if compared with classical HTS compounds; hence, fragment libraries provide a good diversity with a relatively low number of compounds. Classical biochemical assays are often not suitable to detect the low binding affinities involved, so some well known biophysical techniques, such as nuclear magnetic resonance and x-ray, have been opportunely modified in order to render them able to perform the task. When selecting fragments suitable for subsequent optimization, a useful parameter has been introduced, the ligand efficiency, which is defined as the free energy of binding divided by the non-hydrogen atom count. Once selected, a fragment must undergo a heavy elaboration to improve binding affinity, at the same time acquiring drug-like properties. There are two main ways to go on at this point. The most common one is the so-called 'fragment evolution', consisting of a stepwise and systematic addition of chemical functionalities to the starting fragment core, together with a continuous feedback for pharmacological and physicochemical properties. The second one, less common but with great potential, is 'fragment linking': when two or more fragment hits are found to bind in adjacent regions of the target protein, they can be linked through appropriate spacers to rapidly produce a single molecule with much higher binding affinity. Two representative case histories are described: Abbott's ABT 518, an MMP (matrix metalloproteinase) inhibitor, and Eli-Lilly's LY-517717, an inhibitor of factor Xa serine protease. In addition, a list of molecules claimed to be derived from fragment approach and currently undergoing clinical trials is presented.