In crime investigations, geographic profiling involves inferring information about ‘whodunit’ from information about where and when the crime occurred. Such inferences are possible because people usually commit crime in places they know, rather than seeking opportunities elsewhere. The ‘spatial signature’ of a crime location thus reflects the ‘mental map’ of the offender: the locations they are familiar with from their everyday activities, such as where they live, work, go to school or visit family and friends. Previous empirical research suggests that people are more likely to commit crime near some activity locations in their mental map than others. However, these differences have not been accounted for in theory or in geographic profiling methods—particularly methods that infer the most likely suspect to have committed a crime, given the proximity of their activity locations to the crime. This thesis aimed to enhance geographic profiling by developing, validating, and applying a theoretical model that explains how people’s different activity locations influence their crime locations. To achieve this aim, I first drew on existing theoretical and empirical literature to propose a theoretical model that identifies the attributes of people’s activity locations that influence whether they will commit crime nearby and the psychological mechanisms through which this influence occurs. Next, I tested hypotheses derived from the model to add to its empirical support, using New Zealand Police data on approximately 60,000 offenders who committed burglaries, robberies and extra-familial sex offences between 2009 and 2018. This national dataset included details of the offenders’ crime locations and their pre-crime activity locations such as home addresses, family members’ home addresses, schools, and the locations of prior crimes and other police interactions. I then applied the theoretical model in a new geographic profiling algorithm: the Geographic Profiling: Suspect Mapping and Ranking Technique (GP-SMART). For a given crime, GP-SMART predicts suspects’ probability of committing the crime given the proximity of their activity locations to the crime and the attributes of those activity locations that influence whether they will commit crime nearby—as described in the theoretical model—and prioritises the suspects accordingly. I tested GP-SMART’s accuracy by investigating how often it placed the actual offender among the top ranked suspects. The theoretical model proposed that people are more likely to commit crime in locations where they have reliable knowledge that is relevant to the future crime. Attributes of people’s activities—such as their frequency and similarity to the future crime—affect the development of reliable and relevant knowledge. The empirical tests confirmed that offenders were generally more likely to commit crime closer to the kinds of activity locations that people visit more frequently (e.g., home versus family homes) or likely to impart more relevant knowledge about crime opportunities (e.g., prior crimes versus prior victim or witness location). They were also more likely to commit crime near prior activity locations that would generate both reliable and relevant knowledge, than near prior activity locations lower on either or both dimensions. By accounting for the theoretically salient attributes of suspects’ activity locations, GP-SMART ranked the offender at or near the top of the suspect list at rates exceeding chance and those produced by baseline methods—approximating existing algorithms—that use proximity alone. This thesis makes significant theoretical, methodological and practical contributions. The theoretical model explicitly extends crime pattern theory, and the empirical studies add to the evidence base for this extension and core tenets of both the environmental criminology and investigative psychology theoretical approaches to geographic profiling. The police data included a wider range of activity locations than used in past crime location choice studies based on administrative data, and the findings highlight their utility for signalling offenders’ mental maps. Indeed, the size of the dataset required developing a novel method for sampling when using Discrete Spatial Choice Modelling—a major methodological contribution to the growing discrete crime location choice literature. But of most significance in light of the motivation for this thesis, the GP-SMART results demonstrate that enhancing geographic profiling by incorporating the attributes of people’s various activity locations that influence their crime locations, could help solve crime in practice.