This work documents variability within the Union Springs Formation of the Marcellus subgroup across a variety of length scales along the northern margin of the Appalachian Basin. Visual, geochemical, and mechanical testing methods are utilized to describe rock properties spatially and to understand the drivers of change. Three complete columns of Union Springs Formation in Seneca Falls, NY and Marcellus, NY are lithologically characterized, and placed into a microfacies framework. Based on hand sample analysis, thin section petrography, and scanning electron microscopy, rock properties variation is described both across and within facies. The distribution and arrangement of these facies are tracked across studied stratigraphic sections. Facies positions reveal variable depositional conditions across the distal rim of the basin, inclusive of bottom-water oxygenation, inconsistent sediment transport, and other features important to reconstructing the paleodepositional environment. Geochemical data are used to indicate paleodepositional or diagenetic conditions, which can be important for evaluating resource potential or predicting geomechanical properties. With the goal of better understanding the depositional environment, as well as modern inter-facies and intra-facies variability, this study measures chemical compositional variability of lithologically characterized samples via handheld x-ray fluorescence. Exploring the validity of associations of geochemical signals with depositional facies can clarify the predictive power of geochemical analysis in the Union Springs Formation. Analysis of lithologic, chemical, and faunal characteristics informs analysis of the micro-mechanical behavior of these rocks. Procedures for accurately collecting grain-scale mechanical properties of mudrocks through nanoindentation and microindentation are developed. It is hoped that by creating a better understanding of the characteristic deformation behavior of individual constituents of mudstones within a rock matrix, the bulk deformation behavior can be more accurately modeled and predicted.