Alcohol dependence is a psychiatric disorder with a lifetime prevalence of over 10% and a leading cause of morbidity and premature death. A better understanding of the neural mechanisms underlying alcohol dependence to improve prevention, diagnosis and treatment is thus of great societal interest. Recent advancements in the analysis of neuroimaging data based on machine learning have opened new paths to a better quantitative understanding of the disorder. The present habilitation reviews both works with a focus on improving machine learning methodology and empirical works in which machine learning was applied to investigate the neural basis of alcohol dependence. The methodological works advanced several aspects of machine learning in neuroimaging. In particular, they introduced i) a novel classifier (weighted robust distance – WeiRD), which operates parameter-free, computationally efficient and enables a transparent inspection of feature importances, ii) a method to preprocess neuroimaging data based on multivariate noise normalization, which yielded a substantial improvement in classification performance compared to previous the state-of-the-art, and iii) a novel method to reintroduce meaningful graded information into discretized classification accuracies by utilizing classifier decision values. Drawing on a large neuroimaging dataset of alcohol-dependent patients and controls from the LeAD-study (www.lead-studie.de; clinical trial number: NCT01679145), machine learning methods were applied in empirical works to investigate structural and functional alterations in alcohol dependence. Structural damage associated with alcohol dependence were investigated from two conceptually different angles. A first study was aimed at providing the first quantitative evidence for a long-standing hypothesis about the damaging effects of alcohol – the premature aging hypothesis. To this end, a machine learning model was trained on the relationship between grey-matter pattern information and chronological age in a healthy control group and then applied to the sample of alcohol-dependent patients. The predicted ‘brain age’ of patients was found to be was several years higher than their chronological age, thus not only providing quantitative evidence for brain aging in alcohol dependence, but also showing that these aging effects are indeed substantial in relation to the human lifespan. The second study used machine learning to quantify the predictive accuracy of grey-matter pattern information for the diagnosis and a severity measure (lifetime consumption) of alcohol dependence. On average, machine learning models correctly predicted the diagnosis in three of four subjects and accurately estimated the amount of lifetime alcohol consumption. Closer inspection of the prediction model indicated an important role of dorsal anterior cingulate cortex. Comparison with an experienced radiologist, who, like the classifier, was provided with the structural brain scans of the subjects, demonstrated superior performance of computer-based classification and in addition a more effective consideration of demographic information (age and gender). Finally, a third study used functional magnetic resonance imaging to investigate a specific hypothesis about reduced goal-directed learning in alcohol dependence as well as its relation to relapse after detoxification. Computational modelling in combination with machine learning revealed that the interaction of model-based learning and high alcohol expectancies was predictive of diagnosis (patients versus controls) and treatment outcome (abstainers versus relapsers). This finding was paralleled by a signature of model-based learning in medial prefrontal cortex, which was reduced in patients relative to controls and in relapsers relative to abstainers. In sum, the works presented in this habilitation advance machine learning methods for neuroimaging and show that these methods yield novel insights into the neural basis of alcohol dependence. An emerging theme across the three empirical studies on alcohol dependence is the disturbance of executive frontal brain structure and function, supporting a top-down rather than bottom-up view for the aetiology of alcohol dependence.