Per- and polyfluoroalkyl substances (PFASs) encompass a diverse class of anthropogenic chemicals, characterized by the presence of at least one fully fluorinated methyl (–CF₃) or methylene (–CF₂–) group. The carbon-fluorine bond structure confers high chemical resistance and physical stability, which make PFASs uniquely suited for applications that require hydrophobic or lipophobic qualities, including fire-fighting foams, and a variety of consumer products and industrial applications. However, the same characteristics that make PFASs commercially valuable also render them highly persistent in the environment. PFASs have been detected in surface water, groundwater, and the drinking water supplies of millions across the United States, raising concerns due to their association with adverse public health effects and ecological risks. In response to increasing regulations on PFASs, there is a growing need for forensic tools that can attribute PFAS contamination to specific sources in order to understand historical emissions, assign liability, and support site remediation. Recent advances in high-resolution mass spectrometry (HRMS) and machine learning (ML) have made it possible to develop detailed chemical fingerprints characteristic of individual PFAS sources, which can be used to support source allocation efforts in the field. However, there remains limited guidance in the literature on the number and types of PFASs or co-occurring compounds that should be included in these fingerprints, or how environmental processes may alter their integrity in the field. Therefore, the overarching goal of this dissertation is to develop comprehensive PFAS fingerprints for putative PFAS sources and understand how these fingerprints are altered by their fate and transport in the environment. To achieve this, three complementary research projects were conducted. First, I applied target and suspect screening with ML to identify PFAS fingerprints for six PFAS sources in aquatic matrices. High classification accuracy was achieved using a small subset of PFASs, offering practical guidance for future monitoring. Second, I performed a nontarget analysis to identify non-PFAS features that can serve as source-specific markers and developed a ML workflow to interpret the resulting high-resolution data. Findings were organized into a tiered framework of chemical features with increasing annotation confidence, which can support development of targeted HRMS methods for PFAS source differentiation. Third, I simulated environmental processes including dilution, transport, transformation, and source mixing, to evaluate how fingerprint composition and classifier performance change with distance from the source and time since release. Results showed that fingerprint distortion, rather than PFAS loss alone, drove misclassification, revealing process-specific thresholds for forensic reliability. Collectively, this dissertation presents an interpretable, comprehensive framework for PFAS source allocation by means of a fingerprinting approach and identifies the key chemical and environmental conditions under which such forensic tools can be confidently applied.