This research investigates the compositional homogeneity of potash-limesilica glasses from the 12th-17th centuries in Northern Europe, and the significance of this with respect to compositional studies of archaeological glasses. The variables in the glass making process that influence the formation of a homogeneous glass are discussed, and investigated using laboratory replication of beech and bracken ash glasses. The experimental results are compared to archaeological material from glass production sites at Blunden's Wood, Knightons, Sidney Wood, and Little Birches in England, and Hils in Germany. Backscattered scanning electron microscope (SEM) imaging is used to qualify the extent of inhomogeneity in both the experimental and archaeological samples. It is confirmed that visually homogeneous glasses can contain inhomogeneities that are only visible under backscattered SEM imaging. It is seen that the size and orientation of inhomogeneities is varied, and specific glass artefact types (such as crucible and waste glass) are more prone to inhomogeneity than fully formed glass (such as window and vessel glass). Electron microprobe analysis (EPMA) is used to quantify the extent of elemental variations present in the inhomogeneous archaeological glasses. The results show that a number of elements are significantly influenced by inhomogeneity, including those (such as calcium, magnesium and sodium) which are commonly used to form compositional groupings of medieval glass. It is concluded that although a number of variables in the glass making process influence the formation of a homogeneous glass, specific variables, such as increased furnace temperature and a high alkali concentration in the ash, appear to be the dominating factors. The presence of large elemental variations in a number of the archaeological glasses analysed confirms that inhomogeneity is a vital consideration in compositional studies of this material, and that particular care must be exercised when using analytical techniques that require only a small sample size.