The accumulation of knowledge is a key driver of technological change and economic growth. Significant attention has been directed to the processes of knowledge production in a spatial context, but little attention has been given to the type of knowledge produced within specific places. The objectives of the present study are to map the US technology/knowledge space, to examine the evolution of that space over the time period 1975–2005, and to investigate the character of knowledge cores within US cities. The knowledge space is based on the proximity of technology classes, utilizing measures derived from co-classification information contained in patent documents. Results show that over time, patents increasingly cluster within technology classes that are located close to one another in technology space. They also reveal considerable heterogeneity in measures of technological specialization across US metropolitan areas. In general, smaller cities tend to display higher levels of knowledge relatedness, often because invention is controlled by a small number of firms with a limited range of technological know-how. Larger cities generate knowledge that is more broadly dispersed across the US knowledge space. Some cities maintain their technological coherence over time, the technological trajectories of others fracture and dissipate, while yet in other cities new technology cores emerge and develop. Higher levels of technological relatedness (specialization) in cities are linked to faster rates of knowledge production and to distinctive trajectories of knowledge evolution.