How an animal allocates time and energy to different activities has repercussions for its survival, fitness, and response to changing environmental conditions. Fine-scale information on animal behaviour and energy expenditure can help develop better informed, targeted strategies to conserve and manage wild animal populations in the face of increasing anthropogenic impact, and climate change. However, wild animals are notoriously difficult to observe. Miniature data-logging devices attached to animals can address these difficulties by making it possible to remotely observe animals through the eyes of recorded data. Inertial sensors such as accelerometers that are sensitive to movement and the lack of it have offered a powerful tool to remotely quantify both animal behaviour and energetics over the last two decades. However, even though many studies have been able to infer common, coarse-scale animal behaviours such as resting, locomotion, and feeding/foraging from accelerometer data, there is a lack of general methods that can be applied across species and further the uptake of this technology. Further, few studies have been able to resolve fine-scale behaviours, possibly because of the nonstationary signals recorded during complex behaviour. Finally, techniques used to quantify energy expenditure from accelerometer data rely on assumptions that may not always be valid. Magnetometers, another kind of inertial sensor often included in animal tags, have shown the potential to capture both static and dynamic components of movement, as well as the ability to detect some behaviours that accelerometers miss. Yet, no study has explored the potential of magnetometers to recognise common animal behaviours, and perform a one-to-one comparison between accelerometers and magnetometers to assess the relative strengths and weaknesses of these two sensors. The main objectives of this thesis were to address these issues by developing general, robust methods to quantify behaviour and energy expenditure (EE) from inertial sensor data. The thesis first presents a biomechanically driven learning approach for the recognition of common animal behaviours. Using data collected on 10 wild meerkats (Suricata suricatta), I show the high accuracy and robustness of this model to recognise common meerkat behaviours (resting, vigilance, foraging, and running) not only with accelerometers, but also, in a separate study, with magnetometers. Next, the thesis presents a new approach to resolve fine-scale behaviours. Here, complex behaviour is conceptualised as being composed of characteristic impulsive movements (microevents) producing brief shock signals in accelerometer data. Microevents are first detected using appropriate signal processing, and then recognised using robust machine learning. Finally, inspired from approaches adopted in human studies to estimate EE, this thesis proposes a new method for estimating EE based on metabolic equivalent of task (MET), which depends on only behaviour duration, and so avoids problematic assumptions about accelerometer data. Finally, I come full circle by applying these "engineering" developments to the "ecological" question of how differences in individual traits, such as body mass, affect foraging strategy in free-ranging, wild, group-living meerkats, and find that lighter females adapt their foraging strategy to avoid conflict and loss of food to heavier females.