The action of D-serine as a gliotransmitter has attracted considerable attention in recent years and it is now thought that this molecule may play a key role in essential brain functions, such as learning and memory formation1. However, our ability to study D-serine dynamically in brain tissue is limited by the lack of any non-invasive, genetically encodable biosensors, which are available for other transmitters, such as the Förster Resonance Energy Transfer (FRET)-based optical sensor for glutamate2. The central issue that has prevented the construction of a D-serine specific FRET sensor is the absence of any naturally occurring, and specific, D-serine binding proteins. To compound this problem, our ability to engineer specificity into binding proteins is limited, especially in the absence of a high throughput screening/selection method. To overcome these limitations, we have employed computational protein design to guide the engineering of a naturally occurring D-alanine/glycine binding protein (DalS)3 towards increased D-serine specificity. The inclusion of this binding domain in a FRET-sensor created the first optical biosensor specific for D-serine. Several iterations of computational design and experimental characterisation yielded two extremely thermostable FRET biosensors, the first of which bound D-serine with ~ 40-fold higher specificity than the wild type (relative to the competing ligand, glycine), and the second of which bound D-serine with very high affinity (Kd = 7 µM). Together, these sensors highlight the power of computational design and represent the first sensors of this class. It is hoped that they will now become widely used experimental tools that could yield new insight into the role of D-serine in the brain.