A connectionist model of harmonic resolution, based on affective principles, is proposed. The connectionist model consists of a note-recognition layer; and a chord-recognition layer; the chord- recognition layer feeds back to the note-recognition layer to produce a resonance loop. Affect is measured by the degree of mutual maintenance of competing units. This measure is derived from the aesthetic principle of unity in diversity. The affective measure permits the rules of harmony to be understood on aesthetic grounds, rather than as a loose collection of observed regularities. It is also shown how the model is broadly consistent with the rules of common harmonic practice and, in addition, is capable of making predictions about dynamic and adaptive effects. Simulations are performed on two types of resolution: the resolution of a nonharmonic note and cadential resolution. The former set of simulations includes the anchoring of nonharmonic notes, the asymmetry of nonharmonic resolution, and the nonharmonic support mechanisms of emphasis and the suspension. The last set includes voice leading in the cadence, the dominant seventh, the French sixth, and a longer cadence. The paper concludes with a discussion of how some expectation-based effects could be included in a modified model that retains the same affective measure.