In this paper, I argue that the folk ballad tradition of the Faroe Islands, to date never examined by metrists, is best accounted for by a constraint-based metrical grammar (Kiparsky 2006, Hayes et al. 2012) that generates a meter far closer to the actual data encountered than the iterative rule-based account of Fabb and Halle (2008), which both misses generalisations and predicts unattested line types. The ballad performances also reveal a general correspondence between strong metrical positions, strong dance steps and strong musical beats, indicating that rhythm is largely determined by metrical prominence. In support of my conclusions I draw upon both the ballad texts and audio/video recordings of sections of sample ballads I made on the Faroe Islands. I propose that an inventory of constraints on the correspondence between a metrical template and a syllabic representation accurately captures the attested lines, and also accounts for generalisations about stress-meter alignment. By contrast, Fabb and Halle’s deletion rules cannot exclude nonexistent lines without additional stipulations. Where my analysis precludes unattested lines by prohibiting lexical stress in weak position, Fabb and Halle must stipulate that such lines have ill-formed metrical grids. To do so, they must posit ad hoc restrictions on their deletion rules, which must not apply in other cases lest they exclude attested line types. Moreover, generalisations which seem to hold of structural boundaries in the stanza are easily captured via constraints on metrical position; such facts are lost on Fabb and Halle’s proposal unless they combine constraints with rules, leading to redundancy which my approach avoids. From recordings I found that although there is not always a one-to-one mapping from meter to rhythm, evidence from the ringdance shows that metrical and musical peaks tend to align. With the exception of refrains and Danish-language ballads, which exhibit some irregularity, (2) is unviolated in my recording sample: (2) Strong dance steps will always correspond to strong metrical positions. (2) is not bidirectional, in that not all strong positions receive strong steps; nevertheless, it accurately predicts where a strong position will occur in the stanzas. On my account, this fact about text-setting provides a source of secondary evidence for a given parse. Therefore, I propose that a constraint-based approach to meter captures Faroese data accurately, rules out unattested lines, avoids redundancy and sheds light on the relation between meter and performance.