Finding novel compounds as starting points for optimization is a major challenge in drug discovery research. Fragment-based methods have emerged in the past ten years as an effective way to sample chemical diversity with a limited number of low molecular weight compounds. The structures of the fragments(s) binding to the protein can then be used to design new compounds with increased affinity, specificity and novelty. This article describes the Vernalis approach to fragment based drug discovery, called SeeDs (Structural exploitation of experimental Drug startpoints). The approach includes the design of a fragment library, identification of fragments that bind competitively to a target by ligand-based NMR techniques and protein crystal structures to characterize binding. Fragments that bind are then evolved to hits, either by growing the fragment or by combining structural features from a number of compounds. The process is illustrated with examples from recent medicinal chemistry programmes to discover compounds against the oncology targets Hsp90 and PDK1. In addition, we summarise our experience with using molecular docking calculations to predict fragment binding and anecdotes on the selectivity and binding modes for fragments seen against a range of targets.